CN103956170B - A kind of eliminate the method for reverberation, device and equipment - Google Patents

Abstract

Embodiments provide and a kind of eliminate the method for reverberation, device and equipment, it is possible to accurately eliminate the reverb signal in voice signal.The method includes: the energy of the 1st frame of the continuous L frame voice signal of acquisition i-th and 1 voice-input device input of other M to the voice signal of L-th frame, 1st frame is to the reverberation energy coefficient of the clean speech signal of L 1 frame and the coefficient of the speech signal energy of L-th frame, and the 1st frame of i-th voice-input device is to the energy of the clean speech signal of L 1 frame；The coefficient of the reverberation energy coefficient of the 1st frame of described i-th voice-input device to the clean speech signal of L 1 frame and the speech signal energy of L-th frame is updated；Estimate that the 1st frame is to reverberation energy to the reverb signal that described L-th frame voice signal produces of the clean speech signal of L 1 frame；Obtain the energy of the clean speech signal of described L-th frame signal.The method is applicable to Audiotechnica field.

Description

A kind of eliminate the method for reverberation, device and equipment

Technical field

The present invention relates to Audiotechnica field, particularly relate to a kind of eliminate the method for reverberation, device and equipment.

Background technology

Sound wave, when indoor propagation, will be reflected by barriers such as wall, ceiling, floors, often reflect and once will be hindered Thing is hindered to absorb.So, after sound source stops sounding, sound wave the most just to disappear through multiple reflections and absorption in indoor, We just feel that sound source stops sound after sounding and also continues to a period of time, and this phenomenon is called reverberation, is called mixed during this period of time The time of sound.

Under some scenes, such as in conference system, the impact of reverberation can cause the decline of speech intelligibility.In order to carry Rise Consumer's Experience, need reverberation is suppressed, to improve the definition of voice.

Prior art proposes the following two kinds solution for this problem:

The first scheme: be analyzed the phase place of the signal that mike obtains in time domain, estimates adaptive-filtering system Number, signal is filtered by the adaptive-filtering coefficient obtained by estimation, the signal after the reverberation that is eliminated.

But, adopt the impact being easily subject to sound source in this way with mike distance, such as people is when speaking, along with people Health, the movement of head, sound source also can change with the distance of mike, so, the phase place of signal that mike obtains Also changing, it is inaccurate for causing the adaptive-filtering coefficient estimated.

First scheme: remove reverberation based on statistical model at frequency domain, estimates the reverberation of former frame speech signal energy Energy coefficient, thus estimate the reverberation energy in the speech energy of present frame, but, the former frame obtained based on statistical model The reverberation energy coefficient of speech signal energy may be inaccurate, and the effect causing dereverberation is not ideal enough.

Summary of the invention

Embodiments of the invention provide a kind of methods, devices and systems eliminating reverberation, it is possible to eliminate voice letter accurately Reverb signal in number, obtains pure voice signal.

For reaching above-mentioned purpose, embodiments of the invention adopt the following technical scheme that

First aspect, embodiments provides a kind of method eliminating reverberation, and the method includes:

Obtain the 1st frame of continuous L frame voice signal of i-th and other M-1 voice-input device input to L-th frame The energy of voice signal, and the continuous L frame voice signal of described i-th and other M-1 voice-input device input 1st frame to the reverberation energy coefficient of the clean speech signal of L-1 frame and the coefficient of the speech signal energy of L-th frame, and 1st frame of the continuous L frame voice signal of i-th voice-input device input to the energy of the clean speech signal of L-1 frame, its In, described voice signal comprises clean speech signal and reverb signal, and described continuous L frame voice signal is frequency-region signal, described L More than 1；

The 1st frame according to described i-th and the continuous L frame voice signal of other M-1 voice-input device input is extremely The energy of the voice signal of L-th frame, and the continuous L frame voice letter of described i-th and other M-1 voice-input device input Number the 1st frame to the reverberation energy coefficient of the clean speech signal of L-1 frame and the coefficient of the speech signal energy of L-th frame, Clean speech signal to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to L-1 frame The coefficient of the speech signal energy of reverberation energy coefficient and L-th frame is updated；

1st frame of the continuous L frame voice signal according to the input of described i-th voice-input device is pure to L-1 frame The 1st frame after the energy of voice signal, and the renewal of described i-th voice-input device is to the clean speech signal of L-1 frame Reverberation energy coefficient, estimate that the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs is to L-1 frame The clean speech signal reverberation energy to the reverb signal that described L-th frame voice signal produces；

The energy of L-th frame signal of the continuous L frame voice signal according to the input of described i-th voice-input device and described Reverberation energy, it is thus achieved that the energy of the clean speech signal of described L-th frame signal.

In the embodiment that the first is possible, in conjunction with first aspect, described acquisition i-th and other M-1 voice Input equipment input continuous L frame voice signal the 1st frame to the clean speech signal of L-1 frame reverberation energy coefficient and The coefficient of the speech signal energy of L-th frame specifically includes:

The slope q of the Energy Decay Curve according to default impulse signal, by the first formula, it is thus achieved that described i and 1st frame of the continuous L frame voice signal of other M-1 voice-input device inputs is to clean speech signal mixed of L-1 frame Ring energy coefficient g₁、g₂…g_L-1And the coefficient g of the speech signal energy of L-th frame_L, wherein, described first formula is:

Wherein, 1 ＜=l ＜=L.

In the embodiment that the second is possible, in conjunction with first aspect or embodiment that the first is possible, described L is mixed Sound time T rounds with the ratio of the predetermined time period of frame, wherein, and described default impulse letter according to described reverberation time T Number the slope q of Energy Decay Curve, the time of the energy attenuation of the described impulse signal of estimation to preset energy.

In the embodiment that the third is possible, can in conjunction with first aspect, the first possible embodiment or the second Can embodiment in any one, to described i-th voice-input device input continuous L frame voice signal the 1st frame extremely The reverberation energy coefficient of the clean speech signal of L-1 frame is updated specifically including:

The energy of the continuous L frame signal of i-th voice-input device and jth voice-input device is obtained according to the second formula The difference of amount, wherein, described second formula is:

Wherein, as 1 ＜=l ＜=L-1, described g_j,lE_i,lDefeated for jth voice Enter the reverberation energy coefficient of the clean speech signal of the l frame of equipment and the signal energy of the l frame of i-th voice-input device Convolution, described g_i,lE_j,lThe reverberation energy coefficient of clean speech signal of l frame and jth for i-th voice-input device The convolution of the signal energy of the l frame of individual voice-input device, as l=L, described g_j,lE_i,lFor jth voice-input device The convolution of signal energy of L-th frame of coefficient and i-th voice-input device of speech signal energy of L-th frame, described g_{i, l}E_j,lThe L-th frame of coefficient and jth voice-input device of speech signal energy for the L-th frame of i-th voice-input device The convolution of signal energy；

Described i-th voice-input device is obtained with other M-1 voice-input device two-by-two according to described second formula Between the difference of energy of continuous L frame signal；

According to described i-th voice-input device and other M-1 voice-input device continuous L frame signal between any two The difference of energy, it is thus achieved that three formula, described 3rd formula is:

Wherein, as 1 ＜=l ＜=L-1, describedPure for i-th voice-input device l frame The updated value of the reverberation energy coefficient of clean voice signal, described g_i,lFor i-th voice-input device l frame clean speech signal Reverberation energy coefficient, described μ_lThe renewal coefficient of the l frame for pre-seting, described h (e_i, E_l) it is according to i-th phonetic entry The difference of the energy of equipment and other M-1 voice-input device continuous L frame signal between any two, and other M-1 voice The value that 1st frame of the continuous L frame voice signal of input equipment to the energy of the voice signal of L-th frame obtains；

According to described 3rd formula, it is thus achieved that the 1st of the described continuous L frame voice signal of described i-th voice-input device Frame is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the updated value of the coefficient of the speech signal energy of L-th frame.

In the 4th kind of possible embodiment, in conjunction with the embodiment that the third is possible, described μ_l＞=μ_l-1, wherein, Described μ_lThe renewal coefficient of the l frame for pre-seting, described μ_l-1The renewal coefficient of the l-1 frame for pre-seting, described l-1 Frame is the former frame of l frame, described 1 ＜=l-1.

Second aspect, embodiments provides a kind of method eliminating reverberation, and the method includes:

Obtain the 1st frame of continuous L frame voice signal of i-th and other M-1 voice-input device input to L-th frame The amplitude of voice signal, and the continuous L frame voice signal of described i-th and other M-1 voice-input device input 1st frame to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and the coefficient of the clean speech signal amplitude of L-th frame, And the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to the width of the clean speech signal of L-1 frame Degree, wherein, described voice signal comprises clean speech signal and reverb signal, and described continuous L frame voice signal is frequency-region signal, Described L is more than 1；

The 1st frame according to described i-th and the continuous L frame voice signal of other M-1 voice-input device input is extremely The amplitude of the voice signal of L-th frame, and the continuous L frame voice letter of described i-th and other M-1 voice-input device input Number the 1st frame to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and L-th frame clean speech signal amplitude be Number, to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to the clean speech signal of L-1 frame Reverberation amplitude coefficient and the coefficient of clean speech signal amplitude of L-th frame be updated；

1st frame of the continuous L frame voice signal according to the input of described i-th voice-input device is pure to L-1 frame The 1st frame after the amplitude of voice signal, and the renewal of described i-th voice-input device is to the clean speech signal of L-1 frame Reverberation amplitude coefficient, estimate that the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs is to L-1 frame The clean speech signal reverberation amplitude to the reverb signal that described L-th frame voice signal produces；

The amplitude of L-th frame signal of the continuous L frame voice signal according to the input of described i-th voice-input device and described Reverberation amplitude, it is thus achieved that the amplitude of the clean speech signal of described L-th frame signal.

The third aspect, embodiments provides a kind of device eliminating reverberation, and this device includes: first obtains list Unit, the first updating block, the first evaluation unit and second obtains unit；

Described first obtains unit, for obtaining i-th and the continuous L frame of other M-1 voice-input device input 1st frame of voice signal to the energy of the voice signal of L-th frame, and described i-th and other M-1 voice-input device defeated 1st frame of the continuous L frame voice signal entered is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the language of L-th frame The coefficient of tone signal energy, and the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to L-1 frame The energy of clean speech signal, wherein, described voice signal comprises clean speech signal and reverb signal, described continuous L frame language Tone signal is frequency-region signal, and described L is more than 1；

Described first updating block, individual for the described i-th and other M-1 obtaining unit transmission according to described first 1st frame of the continuous L frame voice signal of voice-input device input to the energy of the voice signal of L-th frame, and described i-th with And the 1st frame of the continuous L frame voice signal of other M-1 voice-input device inputs is to the clean speech signal of L-1 frame The coefficient of the speech signal energy of reverberation energy coefficient and L-th frame, the continuous L to the input of described i-th voice-input device 1st frame of frame voice signal is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the speech signal energy of L-th frame Coefficient be updated；

Described first evaluation unit, for obtaining, according to described first, the described i-th voice-input device that unit sends 1st frame of the continuous L frame voice signal of input is to the energy of the clean speech signal of L-1 frame, and described first updating block The 1st frame after the renewal of the described i-th voice-input device sent is to the reverberation energy system of the clean speech signal of L-1 frame Number, estimates the 1st frame clean speech letter to L-1 frame of the continuous L frame voice signal that described i-th voice-input device input Number reverberation energy of reverb signal that described L-th frame voice signal is produced；

Described second obtains unit, for obtaining, according to described first, the described i-th voice-input device that unit sends The described reverberation energy that the energy of the L-th frame signal of the continuous L frame voice signal of input and described first evaluation unit send, obtains Obtain the energy of the clean speech signal of described L-th frame signal.

In the embodiment that the first is possible, in conjunction with the third aspect, described first obtain unit specifically for:

The slope q of the Energy Decay Curve according to default impulse signal, by the first formula, it is thus achieved that described i and 1st frame of the continuous L frame voice signal of other M-1 voice-input device inputs is to clean speech signal mixed of L-1 frame Ring energy coefficient g₁、g₂…g_L-1And the coefficient g of the speech signal energy of L-th frame_L, wherein, described first formula is:

Wherein, 1 ＜=l ＜=L.

In the embodiment that the second is possible, in conjunction with the third aspect or embodiment that the first is possible, described L is mixed Sound time T rounds with the ratio of the predetermined time period of frame, wherein, and described default impulse letter according to described reverberation time T Number the slope q of Energy Decay Curve, the time of the energy attenuation of the described impulse signal of estimation to preset energy.

In the embodiment that the third is possible, can in conjunction with the third aspect, the first possible embodiment or the second Can embodiment in any one, described first updating block specifically for:

The energy of the continuous L frame signal of i-th voice-input device and jth voice-input device is obtained according to the second formula The difference of amount, wherein, described second formula is:

Wherein, as 1 ＜=l ＜=L-1, described g_j,lE_i,lDefeated for jth voice Enter the reverberation energy coefficient of the clean speech signal of the l frame of equipment and the signal energy of the l frame of i-th voice-input device Convolution, described g_i,lE_j,lThe reverberation energy coefficient of clean speech signal of l frame and jth for i-th voice-input device The convolution of the signal energy of the l frame of individual voice-input device, as l=L, described g_j,lE_i,lFor jth voice-input device The convolution of signal energy of L-th frame of coefficient and i-th voice-input device of speech signal energy of L-th frame, described g_{i, l}E_j,lThe L-th frame of coefficient and jth voice-input device of speech signal energy for the L-th frame of i-th voice-input device The convolution of signal energy；

Described i-th voice-input device is obtained with other M-1 voice-input device two-by-two according to described second formula Between the difference of energy of continuous L frame signal；

According to described i-th voice-input device and other M-1 voice-input device continuous L frame signal between any two The difference of energy, it is thus achieved that three formula, described 3rd formula is:

Wherein, as 1 ＜=l ＜=L-1, describedPure for i-th voice-input device l frame The updated value of the reverberation energy coefficient of clean voice signal, described g_i,lFor i-th voice-input device l frame clean speech signal Reverberation energy coefficient, described μ_lThe renewal coefficient of the l frame for pre-seting, described h (e_i, E_l) it is according to i-th phonetic entry The difference of the energy of equipment and other M-1 voice-input device continuous L frame signal between any two, and other M-1 voice The value that 1st frame of the continuous L frame voice signal of input equipment to the energy of the voice signal of L-th frame obtains；

According to described 3rd formula, it is thus achieved that the 1st of the described continuous L frame voice signal of described i-th voice-input device Frame is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the updated value of the coefficient of the speech signal energy of L-th frame.

In the 4th kind of possible embodiment, in conjunction with the embodiment that the third is possible, described μ_l＞=μ_l-1, wherein, Described μ_lThe renewal coefficient of the l frame for pre-seting, described μ_l-1The renewal coefficient of the l-1 frame for pre-seting, described l-1 Frame is the former frame of l frame, described 1 ＜=l-1.

Fourth aspect, embodiments provides a kind of device eliminating reverberation, and this device includes: the 3rd obtains list Unit, the second updating block, the second evaluation unit and the 4th acquisition unit:

Described 3rd obtains unit, for obtaining i-th and the continuous L frame of other M-1 voice-input device input 1st frame of voice signal to the amplitude of the voice signal of L-th frame, and described i-th and other M-1 voice-input device defeated 1st frame of the continuous L frame voice signal entered is pure to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and L-th frame The coefficient of clean voice signal amplitude, and the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to L-1 The amplitude of the clean speech signal of frame, wherein, described voice signal comprises clean speech signal and reverb signal, described continuous L Frame voice signal is frequency-region signal, and described L is more than 1；

Described second updating block, individual for the described i-th and other M-1 obtaining unit transmission according to the described 3rd 1st frame of the continuous L frame voice signal of voice-input device input to the amplitude of the voice signal of L-th frame, and described i-th with And the 1st frame of the continuous L frame voice signal of other M-1 voice-input device inputs is to the clean speech signal of L-1 frame The coefficient of the clean speech signal amplitude of reverberation amplitude coefficient and L-th frame, the company to the input of described i-th voice-input device 1st frame of continuous L frame voice signal is to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and the clean speech of L-th frame The coefficient of signal amplitude is updated；

Described second evaluation unit, for obtaining, according to the described 3rd, the described i-th voice-input device that unit sends 1st frame of the continuous L frame voice signal of input is to the amplitude of the clean speech signal of L-1 frame, and described second updating block The 1st frame after the renewal of the described i-th voice-input device sent is to the reverberation amplitude system of the clean speech signal of L-1 frame Number, estimates the 1st frame clean speech letter to L-1 frame of the continuous L frame voice signal that described i-th voice-input device input Number reverberation amplitude of reverb signal that described L-th frame voice signal is produced；

Described 4th obtains unit, for obtaining, according to the described 3rd, the described i-th voice-input device that unit sends The described reverberation amplitude that the amplitude of the L-th frame signal of the continuous L frame voice signal of input and described second evaluation unit send, obtains Obtain the amplitude of the clean speech signal of described L-th frame signal.

5th aspect, embodiments provides a kind of equipment eliminating reverberation, and this equipment includes: first transceiver and First processor；

Described first transceiver is used for: receive i-th voice-input device and other M-1 voice-input device input Continuous L frame voice signal；

Described first processor is used for: obtain i-th and the continuous L frame language of other M-1 voice-input device input 1st frame of tone signal is to the energy of the voice signal of L-th frame, and described i-th and other M-1 voice-input device input The 1st frame of continuous L frame voice signal to the reverberation energy coefficient of the clean speech signal of L-1 frame and the voice of L-th frame The coefficient of signal energy, and pure to L-1 frame of the 1st frame of the continuous L frame voice signal of i-th voice-input device input The energy of clean voice signal, wherein, described voice signal comprises clean speech signal and reverb signal, described continuous L frame voice Signal is frequency-region signal, and described L is more than 1；

The 1st frame according to described i-th and the continuous L frame voice signal of other M-1 voice-input device input is extremely The energy of the voice signal of L-th frame, and the continuous L frame voice letter of described i-th and other M-1 voice-input device input Number the 1st frame to the reverberation energy coefficient of the clean speech signal of L-1 frame and the coefficient of the speech signal energy of L-th frame, Clean speech signal to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to L-1 frame The coefficient of the speech signal energy of reverberation energy coefficient and L-th frame is updated；

1st frame of the continuous L frame voice signal according to the input of described i-th voice-input device is pure to L-1 frame The 1st frame after the energy of voice signal, and the renewal of described i-th voice-input device is to the clean speech signal of L-1 frame Reverberation energy coefficient, estimate that the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs is to L-1 frame The clean speech signal reverberation energy to the reverb signal that described L-th frame voice signal produces；

The energy of L-th frame signal of the continuous L frame voice signal according to the input of described i-th voice-input device and described Reverberation energy, it is thus achieved that the energy of the clean speech signal of described L-th frame signal.

In the embodiment that the first is possible, in conjunction with the 5th aspect, described first processor specifically for:

The slope q of the Energy Decay Curve according to default impulse signal, by the first formula, it is thus achieved that described i and 1st frame of the continuous L frame voice signal of other M-1 voice-input device inputs is to clean speech signal mixed of L-1 frame Ring energy coefficient g₁、g₂…g_L-1And the coefficient g of the speech signal energy of L-th frame_L, wherein, described first formula is:

Wherein, 1 ＜=l ＜=L.

In the embodiment that the second is possible, in conjunction with the 5th aspect or embodiment that the first is possible, described L is mixed Sound time T rounds with the ratio of the predetermined time period of frame, wherein, and described default impulse letter according to described reverberation time T Number the slope q of Energy Decay Curve, the time of the energy attenuation of the described impulse signal of estimation to preset energy.

In the embodiment that the third is possible, can in conjunction with the 5th aspect, the first possible embodiment or the second Can embodiment in any one, described first processor specifically for:

The energy of the continuous L frame signal of i-th voice-input device and jth voice-input device is obtained according to the second formula The difference of amount, wherein, described second formula is:

Wherein, as 1 ＜=l ＜=L-1, described g_j,lE_i,lDefeated for jth voice Enter the reverberation energy coefficient of the clean speech signal of the l frame of equipment and the signal energy of the l frame of i-th voice-input device Convolution, described g_i,lE_j,lThe reverberation energy coefficient of clean speech signal of l frame and jth for i-th voice-input device The convolution of the signal energy of the l frame of individual voice-input device, as l=L, described g_j,lE_i,lFor jth voice-input device The convolution of signal energy of L-th frame of coefficient and i-th voice-input device of speech signal energy of L-th frame, described g_{i, l}E_j,lThe L-th frame of coefficient and jth voice-input device of speech signal energy for the L-th frame of i-th voice-input device The convolution of signal energy；

Described i-th voice-input device is obtained with other M-1 voice-input device two-by-two according to described second formula Between the difference of energy of continuous L frame signal；

According to described i-th voice-input device and other M-1 voice-input device continuous L frame signal between any two The difference of energy, it is thus achieved that three formula, described 3rd formula is:

Wherein, as 1 ＜=l ＜=L-1, describedPure for i-th voice-input device l frame The updated value of the reverberation energy coefficient of clean voice signal, described g_i,lFor i-th voice-input device l frame clean speech signal Reverberation energy coefficient, described μ_lThe renewal coefficient of the l frame for pre-seting, described h (e_i, E_l) it is according to i-th phonetic entry The difference of the energy of equipment and other M-1 voice-input device continuous L frame signal between any two, and other M-1 voice The value that 1st frame of the continuous L frame voice signal of input equipment to the energy of the voice signal of L-th frame obtains；

According to described 3rd formula, it is thus achieved that the 1st of the described continuous L frame voice signal of described i-th voice-input device Frame is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the updated value of the coefficient of the speech signal energy of L-th frame.

In the 4th kind of possible embodiment, in conjunction with the embodiment that the third is possible, described μ_l＞=μ_l-1, wherein, Described μ_lThe renewal coefficient of the l frame for pre-seting, described μ_l-1The renewal coefficient of the l-1 frame for pre-seting, described l-1 Frame is the former frame of l frame, described 1 ＜=l-1.

6th aspect, embodiments provides a kind of equipment eliminating reverberation, it is characterised in that this equipment includes: Second transceiver and the second processor；

Described second transceiver is used for: receive i-th voice-input device and other M-1 voice-input device input Continuous L frame voice signal；

Described second processor is used for: obtain i-th and the continuous L frame language of other M-1 voice-input device input 1st frame of tone signal is to the amplitude of the voice signal of L-th frame, and described i-th and other M-1 voice-input device input Pure to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and L-th frame of the 1st frame of continuous L frame voice signal The coefficient of voice signal amplitude, and the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to L-1 frame The amplitude of clean speech signal, wherein, described voice signal comprises clean speech signal and reverb signal, described continuous L frame Voice signal is frequency-region signal, and described L is more than 1；

The 1st frame according to described i-th and the continuous L frame voice signal of other M-1 voice-input device input is extremely The amplitude of the voice signal of L-th frame, and the continuous L frame voice letter of described i-th and other M-1 voice-input device input Number the 1st frame to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and L-th frame clean speech signal amplitude be Number, to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to the clean speech signal of L-1 frame Reverberation amplitude coefficient and the coefficient of clean speech signal amplitude of L-th frame be updated；

1st frame of the continuous L frame voice signal according to the input of described i-th voice-input device is pure to L-1 frame The 1st frame after the amplitude of voice signal, and the renewal of described i-th voice-input device is to the clean speech signal of L-1 frame Reverberation amplitude coefficient, estimate that the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs is to L-1 frame The clean speech signal reverberation amplitude to the reverb signal that described L-th frame voice signal produces；

The amplitude of L-th frame signal of the continuous L frame voice signal according to the input of described i-th voice-input device and described Reverberation amplitude, it is thus achieved that the amplitude of the clean speech signal of described L-th frame signal.

Embodiments providing and a kind of eliminate the method for reverberation, device and equipment, the method includes: obtain i-th And the 1st frame of the continuous L frame voice signal of other M-1 voice-input device input is to the energy of the voice signal of L-th frame, Pure with the 1st frame of described i-th and the continuous L frame voice signal of other M-1 voice-input device input to L-1 frame The reverberation energy coefficient of clean voice signal and the coefficient of the speech signal energy of L-th frame, and i-th voice-input device is defeated 1st frame of the continuous L frame voice signal entered is to the energy of the clean speech signal of L-1 frame, and wherein, described voice signal comprises Clean speech signal and reverb signal, described continuous L frame voice signal is frequency-region signal, and described L is more than 1；According to described i-th And the 1st frame of the continuous L frame voice signal of other M-1 voice-input device input is to the energy of the voice signal of L-th frame, Pure with the 1st frame of described i-th and the continuous L frame voice signal of other M-1 voice-input device input to L-1 frame The reverberation energy coefficient of clean voice signal and the coefficient of the speech signal energy of L-th frame, to described i-th voice-input device 1st frame of the continuous L frame voice signal of input is to the reverberation energy coefficient of the clean speech signal of L-1 frame and L-th frame The coefficient of speech signal energy is updated；The of continuous L frame voice signal according to the input of described i-th voice-input device 1 frame is to the energy of the clean speech signal of L-1 frame, and the 1st frame after the renewal of described i-th voice-input device is to L- The reverberation energy coefficient of the clean speech signal of 1 frame, estimates the continuous L frame voice letter that described i-th voice-input device inputs Number the 1st frame to reverberation energy to the reverb signal that described L-th frame voice signal produces of the clean speech signal of L-1 frame； The energy of the L-th frame signal of the continuous L frame voice signal according to the input of described i-th voice-input device and described reverberation energy Amount, it is thus achieved that the energy of the clean speech signal of described L-th frame signal.Based on the method, the pure language to the 1st frame to L-1 frame The reverberation energy coefficient of tone signal is updated, and the clean speech signal of the 1st frame obtained to L-1 frame is to described L-th frame The reverberation energy of the reverb signal that signal produces is more accurate, therefore, with it, voice letter can be obtained more accurately The energy of the reverb signal in number, and eliminate the impact of this reverb signal, thus improve voice quality.

Accompanying drawing explanation

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing In having technology to describe, the required accompanying drawing used is briefly described, it should be apparent that, the accompanying drawing in describing below is only this Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to Other accompanying drawing is obtained according to these accompanying drawings.

Fig. 1 is a kind of impulse signal response curve；

Fig. 2 is the Energy Decay Curve of a kind of impulse signal response；

Fig. 3 is a kind of method flow diagram eliminating reverberation；

Fig. 4 is the another kind of method flow diagram eliminating reverberation；

Fig. 5 is the method flow diagram of another elimination reverberation；

Fig. 6 is a kind of device schematic diagram eliminating reverberation；

Fig. 7 is the another kind of device schematic diagram eliminating reverberation；

Fig. 8 is a kind of equipment schematic diagram eliminating reverberation；

Fig. 9 is the another kind of equipment schematic diagram eliminating reverberation.

Detailed description of the invention

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Describe, it is clear that described embodiment is only a part of embodiment of the present invention rather than whole embodiments wholely.Based on Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under not making creative work premise Embodiment, broadly falls into the scope of protection of the invention.

In conjunction with Fig. 1, exemplary describing is a kind of when to a voice-input device, such as to a mike input one Individual impulse signal, during such as an instantaneous speech signal, the time dependent curve of amplitude of signal.In conjunction with Fig. 1, at 0 to 0.02s Input an instantaneous speech signal, without reverb signal, the most only can produce an impulse response in the moment of signal input, The amplitude of signal is 0 afterwards, will not change over time and change, but the reverb signal produced by this impulse signal is mixed Being obtained by mike in the time of sound, the most described mike not only gets described impulse signal, also within the reverberation time always Get the reverb signal of decay always.

Based on this, when to mike one impulse signal of input, the reverb signal produced due to described impulse signal Exist, the impulse response curve that an amplitude as shown in Figure 1 is constantly decayed will be obtained.

Corresponding, in conjunction with Fig. 2, the exemplary Energy Decay Curve describing the impulse signal shown in a kind of Fig. 1.In wink Time voice signal input after, have an energy attenuation by a relatively large margin at once, the energy of signal presents smooth decay afterwards.

When to mike one continuous print voice signal of input, can this continuous print voice signal be divided into some Frame, the voice signal of each frame regards an independent instantaneous speech signal, corresponding one of each independent instantaneous speech signal as Independent Energy Decay Curve, terminates if the reverb signal of an instantaneous speech signal generation needs L frame just can decay, that , the voice signal of L-th frame is produced in L-th frame by the clean speech signal of L-th frame and 1 to L-1 frame instantaneous speech signal Reverb signal forms, and i.e. the energy of the voice signal of L-th frame is by the energy of the clean speech signal of present frame, i.e. L-th frame and front Face the 1st frame is constituted to the energy of the reverb signal of the instantaneous speech generation of L-1 frame.

Concrete, the clean speech signal of L-th frame refers to the direct sound wave that sound source sends in L-th frame, is not comprise reverberation Signal, need the final signal obtained by embodiment of the present invention, it is understood that for the instantaneous speech inputted in L-th frame Signal.

The energy of the voice signal of described L-th frame can be represented by equation below (1):

$E_L=S_L+\underset{l=1}{\overset{L-1}{\mathrm{\Σ}}}{g}_l{S}_l---\left(1\right)$

Wherein, E_LFor the energy of L-th frame voice signal, g_lIt is l frame instantaneous speech signal, i.e. l frame clean speech signal At the energy coefficient of the reverb signal of L-th frame, described S_lIt is the energy of l frame clean speech signal, wherein, 1 ＜=l ＜=L- 1。

In conjunction with Fig. 2, the 1st frame instantaneous speech is to L-th frame, and the impact of the most current reverberation is minimum, the instantaneous language of the 1st frame Sound is g at the reverberation energy that L-th frame produces₁S₁, L-1 frame, namely the instantaneous speech of the former frame of present frame is to present frame The impact of reverberation is maximum, and the instantaneous speech of L-1 frame is g at the reverberation energy that L-th frame produces_L-1S_L-1。

Obviously, g_L-1, g_L-2…g₁It it is descending decay.Concrete, it is believed that g_L-1, g_L-2…g₁It is that equal proportion declines Subtract, it is assumed that g_L-1=a, g_L-2=a², g_L-3=a³…g₁=a^L-1, wherein, a ＜ 1

Based on formula (1), available equation below (2)

$\begin{array}{c}{E}_L=S_L+\underset{l=1}{\overset{L-1}{\mathrm{\Σ}}}{g}_l{S}_l\\ =S_L+{\mathrm{aS}}_{L-1}+a^2{S}_{L-2}+a^3{S}_{L-3}+...+a^{L-1}{S}_1\\ =S_L+a(S_{L-1}+{\mathrm{aS}}_{L-2}+a^2{S}_{L-3}+...+a^{L-2}{S}_1)\\ \≈S_L+{\mathrm{aE}}_{L-1}\end{array}---\left(2\right)$

Concrete, when L takes infinity, above formula (1) can be expressed as formula (3):

E_L=S_L+aE_L-1 (3)

According to coefficient a, and the former frame of L-th frame voice signal, the i.e. energy of the voice signal of L-1 frame, it is possible to obtain Obtain the reverberation energy in the energy of L-th frame voice signal, and then obtain the clean speech energy S of L-th frame_L。

But, based on prior art, it is difficult to obtain accurate a, when a is inaccurate, the effect removing reverberation can be caused The most bad.

In conjunction with Fig. 3, embodiments providing a kind of method eliminating reverberation, the method includes:

301, the 1st frame of continuous L frame voice signal of i-th and other M-1 voice-input device input is obtained extremely The energy of the voice signal of L-th frame, and the continuous L frame voice letter of described i-th and other M-1 voice-input device input Number the 1st frame to the reverberation energy coefficient of the clean speech signal of L-1 frame and the coefficient of the speech signal energy of L-th frame, And the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to the energy of the clean speech signal of L-1 frame Amount, wherein, described voice signal comprises clean speech signal and reverb signal, and described continuous L frame voice signal is frequency-region signal, Described L is more than 1.

At the end of the reverb signal of an instantaneous speech signal generation needs L frame just can decay, a continuous print L frame The energy of voice signal of L-th frame of voice signal by the energy of the clean speech signal of L-th frame and the 1st frame to L-1 frame The energy of reverb signal that produces of instantaneous speech signal constitute, so, the energy of the voice signal of described L-th frame is deducted the 1 frame is to the energy of the reverberation voice signal of L-1 frame, it is possible to obtain the energy of the clean speech signal of described L-th frame.

For example, as in figure 2 it is shown, the Energy Decay Curve corresponding to the impulse response signals shown in Fig. 1, usual feelings Under condition, when the energy of described impulse response signals decays to-60dB from 0dB, it is believed that the energy of described impulse response signals Decay terminates, i.e. when the energy of described impulse response signals decays to the time used by-60dB from 0dB, is described impulse letter Number reverberation time.

Based on this, by calculating the slope q of Energy Decay Curve, it is possible to obtain the reverberation time of described impulse signal, In conjunction with Fig. 2, the Energy Decay Curve corresponding to described impulse response signals can be regarded as a smooth linearity curve, pass through The slope q calculating described Energy Decay Curve can calculate the reverberation time of described impulse response.

If described Energy Decay Curve according to shown in Fig. 2 slope of curve q continue decay, from 0dB decay to-60dB time Between be about 0.32s, i.e. reverberation time be 0.32s.Assume that a frame is 20ms, i.e. 0.02s, the reverberation time of the most described impulse response It is 16 frames.It is to say, the reverb signal of the voice signal of the 16th frame is by 15 frame voice letter before described 16th frame voice signal Reverb signal produced by number is constituted, and the energy of described 16th frame voice signal is by the energy of the clean speech signal of described 16th frame Amount is constituted with the energy of described 1st frame to the reverberation voice signal of the 15th frame.

It should be noted that under same voice environment, such as, in same meeting room, each voice-input device obtains The slope of the Energy Decay Curve of the impulse response of the instantaneous speech signal taken is identical, and same voice-input device is in difference The slope of the Energy Decay Curve of the impulse response that the time obtains also is identical.

For obtaining the slope of the Energy Decay Curve of described impulse response, can be obtained in advance by the method surveyed in advance One impulse response, and obtain its slope q according to the Energy Decay Curve of this impulse response, when phonetic entry starts, permissible Directly utilize described q and be removed the process of reverberation；

An instantaneous speech can also be obtained in several seconds of the beginning of phonetic entry and obtain its Energy Decay Curve, and Obtain the slope q of described Energy Decay Curve；

The Energy Decay Curve of an impulse response can also be obtained by the method for Mathematics structural, and obtain described energy The slope q of attenuation curve.

Concrete, in same voice environment, such as in same meeting room, the q using above-mentioned three kinds of methods to obtain is phase With, in embodiments of the present invention, the Energy Decay Curve obtaining an impulse response by any of the above described a kind of mode is all Feasible, this is not limited by the embodiment of the present invention.

The energy of the 16th frame voice signal is deducted the 1st frame energy to the reverberation voice signal of the 15th frame, it is possible to The energy of the clean speech signal in the 16th frame voice signal.Concrete, due to the instantaneous speech signal of the 1st frame to the 15th frame The energy of reverb signal produced at the 16th frame is not easy to directly obtain, and the reverb signal of the 16th frame is all by the 1st frame to the The instantaneous speech signal of 15 frames, i.e. clean speech signal produce, and based on this, the instantaneous speech signal of the 1st frame to the 15th frame exists The energy of the reverb signal that the 16th frame produces can be obtained by the energy of the clean speech signal of the 1st frame to the 15th.

The reverberation energy coefficient g of the clean speech signal of the 1st frame₁With the clean speech signal energy of the 1st frame the described 1st The convolution of frame, is the reverberation speech signal energy that the instantaneous speech signal of the first frame produces at the 16th frame, thus can use down State the energy that formula 3.1 represents the voice signal of the 16th frame:

$E_{16}=S_{16}+\underset{l=1}{\overset{15}{\mathrm{\Σ}}}{g}_l{S}_l---3.1$

In formula 3.1, E₁₆It is the energy of the 16th frame voice signal, g_lS_lIt it is the reverberation of the clean speech signal of l frame The convolution of the clean speech energy of energy coefficient and l frame, wherein, l=1,2 ..., 15.

According to formula 3.1, obtain the energy of the L-th frame signal of the continuous print L frame signal of voice signal, and the 1st Frame is to the energy of the clean speech signal of L-1 frame, and the 1st frame is to the reverberation energy system of the clean speech signal of L-1 frame Number, it is possible to obtain the energy of clean speech signal in the voice signal of L-th frame.

Concrete, when the energy of signal calculated, signal can be divided into some subbands according to frequency, calculate each respectively The energy of subband, so can reduce amount of calculation, and does not interferes with the accuracy of calculating, certainly, calculates the energy of each subband Being the known technology of those skilled in the art, this is repeated no more by the embodiment of the present invention.

Certainly, the clean speech signal energy of initial some frame voice signals cannot obtain, and can use and estimate The method of meter obtains, such as, when calculating the clean speech signal energy of the 16th frame, and the clean speech letter of the 1st frame to the 15th frame Number energy be now unknown, can be obtained by the method for pre-estimation.

When calculating the clean speech signal energy of the 17th frame, need to use the clean speech signal of the 2nd frame to the 16th frame Energy, at this point it is possible to use the 2nd frame that obtained by the method for pre-estimation to the energy of the clean speech signal of the 15th frame, And the energy of the clean speech signal of the 16th frame that previous step calculates.

When calculating the energy of clean speech signal of the 18th frame, need to use the clean speech letter of the 3rd frame to the 17th frame Number energy, at this point it is possible to use the 3rd frame that obtained by the method for pre-estimation to the energy of the clean speech signal of the 15th frame Amount, and the energy of the clean speech signal of the energy of the clean speech signal of the 16th frame calculated and the 17th frame.

By that analogy, when calculating the energy of clean speech signal of the 31st frame, need the 16th frame used to the 30th frame The energy of clean speech signal be all the energy of the clean speech signal obtained by the present invention program；

That is, if obtaining the reverberation time by the slope of impulse response energy attenuation curve is 16 frames, when Y ＞=31, During the energy of the clean speech signal calculating Y frame, need the energy of the clean speech signal of Y-15 to Y-1 frame used all Calculated by the present invention program and get.

Concrete, it is thus achieved that the energy of L-th frame voice signal and obtain the 1st frame to L-1 frame by the method for pre-estimation The energy of clean speech signal be the known technology of those skilled in the art, this is repeated no more by the embodiment of the present invention.

By formula 3.1, at the energy of the L-th frame signal of the continuous print L frame signal obtaining voice signal, and the 1st frame To the energy of the clean speech signal of L-1 frame, in addition it is also necessary to obtain pure to L-1 frame of the 1st frame of described continuous L frame signal The reverberation energy coefficient g of clean voice signal₁,g₂..., g_L-1, the energy of clean speech signal in L-th frame voice signal could be obtained Amount.

Concrete, following method can be used to obtain the 1st frame when phonetic entry starts to the clean speech of L-1 frame The reverberation energy coefficient g of signal₁,g₂..., g_L-1:

The reverberation energy coefficient g of the clean speech signal of l-1 frame_l-1Reverberation energy with the clean speech signal of l frame Coefficient g_l, and between the slope q of the Energy Decay Curve of impulse response, there is the relation of equation below 3.2:

$q\≈10*\log 10\left(\frac{g_{l-1}}{g_l}\right)---3.2$

OrderThen according to formula 3.2, equation below 3.3 can be obtained:

$F={10}^{\frac{q}{10}}---3.3$

According to F, by equation below 3.4, the reverberation to the clean speech of described L-1 frame of described 1st frame can be obtained Energy coefficient:

g_l=(F)^L-l 3.4

It is not difficult to draw, the reverberation energy coefficient g of the clean speech signal of L-1 frame_L-1=F, is in above-mentioned formula (3) A.

It should be noted that the reverberation energy coefficient because of the 1st frame to the clean speech signal of L-1 frame will not be strict The relation of equal proportion decay, the reverberation energy system of the 1st frame obtained by said method to the clean speech signal of L-1 frame Number g₁,g₂..., g_L-1Likely can be inaccurate, need it is updated.

It is of course also possible to obtain the reverberation to the clean speech signal of L-1 frame of the 1st frame by the method arranging empirical value Energy coefficient g₁,g₂..., g_L-1, this is not limited by the embodiment of the present invention.

Concrete, at the g to i-th voice-input device₁,g₂..., g_L-1During being updated, need to use i-th 1st frame of the continuous L frame voice signal of individual and other M-1 voice-input device inputs is to the energy of the voice signal of L-th frame Amount, and the 1st frame of the continuous L frame voice signal of described i-th and other M-1 voice-input device input is to L-1 frame The reverberation energy coefficient of clean speech signal and the coefficient of speech signal energy of L-th frame, and i-th phonetic entry sets 1st frame of the continuous L frame voice signal of standby input is to the energy of the clean speech signal of L-1 frame.

Certainly, described voice-input device can be mike, it is also possible to one that is conference telephone or video conferencing defeated Enter end, or other voice-input devices；

Other voice-input devices described can be 1, it is also possible to being multiple, this is not limited by the embodiment of the present invention.

It should be noted that the continuous L frame voice signal of i-th voice-input device input and other M-1 voice are defeated Enter the continuous L frame voice signal of equipment input, be the signal that inputs within the same time period (L frame period) of same sound source.

302, according to described i-th and the 1st of the continuous L frame voice signal of other M-1 voice-input device input the Frame is to the energy of the voice signal of L-th frame, and described i-th and the continuous L frame language of other M-1 voice-input device input 1st frame of tone signal to the reverberation energy coefficient of the clean speech signal of L-1 frame and L-th frame speech signal energy be Number, to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to the clean speech signal of L-1 frame Reverberation energy coefficient and the coefficient of speech signal energy of L-th frame be updated.

Concrete, can be according to the continuous L frame voice letter of described i-th and other M-1 voice-input device input Number the 1st frame to the energy of the voice signal of L-th frame, and described i-th and the company of other M-1 voice-input device input 1st frame of continuous L frame voice signal is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the voice signal of L-th frame The coefficient of energy, uses the dual pathways (two voice-input devices) or multichannel (three and above voice-input device) Little mean square algorithm (Least mean square, LMS), the g to i-th voice-input device₁,g₂..., g_L-1Carry out self adaptation Update, it is also possible to by other means to described g₁,g₂..., g_L-1Carry out adaptive updates, concrete, the embodiment of the present invention pair This does not limits.

For example, when phonetic entry starts, during the voice signal carrying out the 16th frame removes reverberation, according to Formula 3.4 obtains the 1st frame reverberation energy coefficient g to the clean speech signal of 1-15 frame₁,g₂..., g₁₅Initial value, so Pass through the dual pathways or multichannel least mean square algorithm afterwards to g₁,g₂..., g₁₅Carry out adaptive updates, it is thus achieved that g₁,g₂..., g₁₅ Updated value.

During the voice signal carrying out the 17th frame removes reverberation, according to the g of above-mentioned acquisition₁,g₂..., g₁₅More New value is as the initial value of the 2nd frame to the reverberation energy coefficient of the clean speech signal of the 16th frame, and passes through the dual pathways or manifold The least mean square algorithm in road is to g₁,g₂..., g₁₅Updated value again update.

By that analogy, order 1 frame reverberation energy coefficient to L-1 frame clean speech signal is all carried out at each frame The process of adaptive updates, and, during the voice signal carrying out L+1 frame removes reverberation, need to use L-th frame language The coefficient of tone signal energy, based on this, is being carried out more the reverberation energy coefficient of the 1st frame to the clean speech signal of L-1 frame During Xin, it is also desirable to the coefficient of L-th frame speech signal energy is updated.

It should be noted that pass through the present invention program, it is assumed that the reverberation time is L frame, it is impossible to remove phonetic entry and start most Continuous L frame voice signal the 1st to the reverb signal in the voice signal of L-1 frame, but, the 1st frame to L-1 frame time Between the ofest short duration, only about 0.3s, whole removing can't be caused too much influence except the effect of reverberation.

303, the 1st frame of the continuous L frame voice signal inputted according to described i-th voice-input device is to L-1 frame The 1st frame after the energy of clean speech signal, and the renewal of described i-th voice-input device is to the clean speech of L-1 frame The reverberation energy coefficient of signal, estimates that the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs is to The clean speech signal of the L-1 frame reverberation energy to the reverb signal that described L-th frame voice signal produces.

The 1st frame of continuous L frame voice signal of i-th voice-input device input is calculated to L-1 according to formula 3.4 The reverberation energy coefficient of the clean speech signal of frame, and the reverberation energy to described 1st frame to the clean speech signal of L-1 frame After coefficient is updated, according to pure to L-1 frame of the 1st frame of the continuous L frame signal of described i-th voice-input device input The energy of clean voice signal, and the 1st frame after described renewal is to the reverberation energy coefficient of the clean speech signal of L-1 frame, root According to Section 2 on the right of the equation of formula (1), it is possible to estimate described 1st frame to L-1 frame clean speech signal to described L The reverberation energy of the reverb signal that frame signal produces.

304, according to described i-th voice-input device input continuous L frame voice signal L-th frame signal energy and Described reverberation energy, it is thus achieved that the energy of the clean speech signal of described L-th frame signal.

The energy of the L-th frame signal of the continuous L frame voice signal inputted by described i-th voice-input device deducts described 1st frame is to reverberation energy to the reverb signal that described L-th frame signal produces of the clean speech signal of L-1 frame, it is possible to Energy to the clean speech signal of described L-th frame voice signal.

The 1st frame compared to the continuous L frame voice signal of the i-th voice-input device input obtained by formula 3.4 To the reverberation energy coefficient of the clean speech signal of L-1 frame, by the described i-th voice-input device after adaptive updates 1st frame of the continuous L frame voice signal of input is to the reverberation energy coefficient of the clean speech signal of L-1 frame, with the actual the 1st Frame to the clean speech signal of L-1 frame reverberation energy coefficient closer to.

Based on this, by the continuous L frame voice signal of the described i-th voice-input device input after adaptive updates 1st frame is to the reverberation energy coefficient of the clean speech signal of L-1 frame, with the continuous L of described i-th voice-input device input 1st frame of frame signal is to the energy of the clean speech signal of L-1 frame, the clean speech signal of the 1st frame obtained to L-1 frame More accurate to the reverberation energy of the reverb signal that described L-th frame signal produces, the effect removing reverberation is more preferable.

Embodiments providing a kind of method eliminating reverberation, the method includes: obtain i-th and other M-1 1st frame of the continuous L frame voice signal of individual voice-input device input is to the energy of the voice signal of L-th frame, and described i-th And the 1st frame of the continuous L frame voice signal of other M-1 voice-input device input is to the clean speech signal of L-1 frame Reverberation energy coefficient and the coefficient of speech signal energy of L-th frame, and the continuous L of i-th voice-input device input 1st frame of frame voice signal is to the energy of the clean speech signal of L-1 frame, and wherein, described voice signal comprises clean speech Signal and reverb signal, described continuous L frame voice signal is frequency-region signal, and described L is more than 1；According to described i-th and other 1st frame of the continuous L frame voice signal of M-1 voice-input device input to the energy of the voice signal of L-th frame, and described the 1st frame of the continuous L frame voice signal of i and other M-1 voice-input devices input is to the clean speech letter of L-1 frame Number reverberation energy coefficient and the coefficient of speech signal energy of L-th frame, the company to the input of described i-th voice-input device 1st frame of continuous L frame voice signal is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the voice signal of L-th frame The coefficient of energy is updated；1st frame of the continuous L frame voice signal according to the input of described i-th voice-input device is to the The 1st frame after the energy of the clean speech signal of L-1 frame, and the renewal of described i-th voice-input device is pure to L-1 frame The reverberation energy coefficient of clean voice signal, estimates the 1st of the continuous L frame voice signal that described i-th voice-input device inputs the Frame is to reverberation energy to the reverb signal that described L-th frame voice signal produces of the clean speech signal of L-1 frame；According to described The energy of the L-th frame signal of the continuous L frame voice signal of i-th voice-input device input and described reverberation energy, it is thus achieved that institute State the energy of the clean speech signal of L-th frame signal.Based on the method, mixed to the clean speech signal of the 1st frame to L-1 frame Ringing energy coefficient to be updated, described L-th frame signal is produced by the clean speech signal of the 1st frame obtained to L-1 frame The reverberation energy of reverb signal is more accurate, therefore, with it, the reverberation in voice signal can be obtained more accurately The energy of signal, and eliminate the impact of this reverb signal, thus improve voice quality.

Embodiments providing a kind of method eliminating reverberation, in conjunction with Fig. 4, the method includes:

401, the 1st frame of continuous L frame voice signal of i-th and other M-1 voice-input device input is obtained extremely The energy of the voice signal of L-th frame, and the continuous L frame voice letter of described i-th and other M-1 voice-input device input Number the 1st frame to the reverberation energy coefficient of the clean speech signal of L-1 frame and the coefficient of the speech signal energy of L-th frame, And the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to the energy of the clean speech signal of L-1 frame Amount, wherein, described voice signal comprises clean speech signal and reverb signal, and described continuous L frame voice signal is frequency-region signal, Described L is more than 1.

The detailed implementation process of step 401 can be found in the step 301 in above-described embodiment, and the embodiment of the present invention is to this no longer Repeat.

402, i-th voice-input device and the continuous L frame signal of jth voice-input device are obtained according to the second formula The difference of energy, wherein, described second formula is:

Wherein, as 1 ＜=l ＜=L-1, described g_j,lE_i,lDefeated for jth voice Enter the reverberation energy coefficient of the clean speech signal of the l frame of equipment and the signal energy of the l frame of i-th voice-input device Convolution, described g_i,lE_j,lThe reverberation energy coefficient of clean speech signal of l frame and jth for i-th voice-input device The convolution of the signal energy of the l frame of individual voice-input device, as l=L, described g_j,lE_i,lFor jth voice-input device The convolution of signal energy of L-th frame of coefficient and i-th voice-input device of speech signal energy of L-th frame, described g_{i, l}E_j,lThe L-th frame of coefficient and jth voice-input device of speech signal energy for the L-th frame of i-th voice-input device The convolution of signal energy.

Concrete, by the dual pathways (two voice-input devices) or multichannel, (two or more voice-input device inputs Voice signal) least mean square algorithm, the reverberation energy system to described 1st frame to the clean speech signal of described L-1 frame The process that number is updated is as follows:

The voice signal of 1 to L-th frame of voice signal due to the continuous L frame of i-th voice-input device input Energy, the reverberation energy coefficient of the 1st frame to the clean speech signal of L-1 frame and the coefficient of the speech signal energy of L-th frame, The energy of the voice signal of 1 to the L-th frame of the voice signal of the continuous L frame of jth voice-input device input, the 1st frame is extremely Between the coefficient of the reverberation energy coefficient of the clean speech signal of L-1 frame and the speech signal energy of L-th frame, exist as follows State the relation of formula 4.1:

$\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{g}_{j,l}{E}_{i,l}=\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{g}_{i,l}{E}_{j,l}---4.1$

Wherein, as 1 ＜=l ＜=L-1, g_j,lE_i,lL frame pure for the signal of jth voice-input device input The convolution of the speech energy of the l frame of the signal that the reverberation energy coefficient of clean voice signal inputs with i-th voice-input device, g_i,lE_j,lThe reverberation energy coefficient of clean speech signal of l frame of signal and jth for the input of i-th voice-input device The convolution of the speech energy of the l frame of the signal of voice-input device input；As l=L, g_j,lE_i,lFor jth phonetic entry The L-th frame of the signal that the coefficient of the speech signal energy of the L-th frame of the signal of equipment input inputs with i-th voice-input device The convolution of speech energy, g_i,lE_j,lFor i-th voice-input device input signal L-th frame speech signal energy be The convolution of the speech energy of the L-th frame of the signal that number inputs with jth voice-input device.

It should be noted that the signal of the continuous print L frame of two voice-input device inputs must be to be sent out by same sound source The voice signal of the continuous L frame within the same time period gone out.

Based on formula 4.1, formula 4.2 can be constructed as follows, be used for representing voice-input device i and voice-input device j The difference of energy of continuous L frame signal:

$e_{i,j}=\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{g}_{j,l}{E}_{i,l}-\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{g}_{i,l}{E}_{j,l}---4.2$

When pure to L-1 frame of the 1st frame of voice signal of continuous L frame of described i-th voice-input device input The reverberation energy coefficient of voice signal and the coefficient of the speech signal energy of L-th frame and described jth voice-input device are defeated 1st frame of the voice signal of the continuous L frame entered is to the reverberation energy coefficient of the clean speech signal of L-1 frame and L-th frame When the coefficient of speech signal energy makes the value minimum of above-mentioned formula 4.2, the continuous L of the described voice-input device i input obtained 1st frame of the voice signal of frame is to the reverberation energy coefficient of the clean speech signal of L-1 frame and voice-input device j input 1st frame of the voice signal of L frame is to the reverberation energy coefficient of the clean speech signal of L-1 frame closest to practical situation continuously, In the continuous L frame voice signal of the voice-input device i obtained by it and voice-input device j input, the 1st frame is to L-1 frame The energy of the reverb signal that clean speech signal produces in L-th frame and the 1st actual frame to L-1 frame clean speech signal are the The error of the energy of the reverb signal that L frame produces is minimum.

403, described i-th voice-input device and other M-1 voice-input device are obtained according to described second formula The difference of the energy of continuous L frame signal between any two.

With the dual pathways, the least mean square algorithm of the voice signal that i.e. two voice-input devices are simultaneously entered same sound source is Example, it is assumed that voice-input device 1 and voice-input device 2 are simultaneously entered the voice signal of same sound source, take in the same time period The continuous L frame signal of the voice signal of voice-input device 1 input and the continuous L of the voice signal of voice-input device 2 input Frame signal.

The difference of the energy of the continuous L frame signal of voice-input device 1 and voice-input device 2 can use equation below 4.3 table Show:

$e_{\mathrm{1,2}}=\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{g}_{2,l}{E}_{1,l}-\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{g}_{1,l}{E}_{2,l}---4.3$

When pure to L-1 frame of the 1st frame of voice signal of continuous L frame of described 1st voice-input device input The reverberation energy coefficient of voice signal and the coefficient of the speech signal energy of L-th frame and described 2nd voice-input device are defeated 1st frame of the voice signal of the continuous L frame entered is to the reverberation energy coefficient of the clean speech signal of L-1 frame and L-th frame When the coefficient of speech signal energy makes the value minimum of above-mentioned formula 4.2, the continuous L of described voice-input device 1 input obtained 1st frame of the voice signal of frame is to the reverberation energy coefficient of the clean speech signal of L-1 frame and voice-input device 2 input 1st frame of the voice signal of L frame is to the reverberation energy coefficient of the clean speech signal of L-1 frame closest to practical situation continuously, In the continuous L frame voice signal of the voice-input device 1 obtained by it and voice-input device 2 input, the 1st frame is to L-1 frame The energy of the reverb signal that clean speech signal produces in L-th frame and the 1st actual frame to L-1 frame clean speech signal are the The error of the energy of the reverb signal that L frame produces is minimum.

The most for example, with triple channel, i.e. three voice-input devices are simultaneously entered the voice signal of same sound source As a example by little mean square algorithm, it is assumed that voice-input device 1, voice-input device 2 and voice-input device 3 are simultaneously entered same sound source Voice signal, take the continuous L frame signal of voice signal of voice-input device 1 input, phonetic entry in the same time period and set The continuous L frame signal of the voice signal of standby 2 inputs and the continuous L frame signal of the voice signal of voice-input device 3 input.

The energy of voice-input device 1, voice-input device 2 and voice-input device 3 continuous L frame signal between any two Difference can use equation below 4.4-4.6 to represent:

$e_{\mathrm{1,2}}=\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{g}_{2,l}{E}_{1,l}-\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{g}_{1,l}{E}_{2,l}---4.4$

$e_{\mathrm{2,3}}=\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{g}_{3,l}{E}_{2,l}-\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{g}_{2,l}{E}_{3,l}---4.5$

$e_{3,1}=\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{g}_{1,l}{E}_{3,l}-\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{g}_{3,l}{E}_{1,l}---4.6$

As described e_1,2, described e_2,3With described e_3,1Time minimum, the continuous L frame of described voice-input device 1 input obtained Voice signal the 1st frame to the clean speech signal of L-1 frame reverberation energy coefficient, voice-input device 2 input company 1st frame of the voice signal of continuous L frame is defeated to reverberation energy coefficient and the voice-input device 3 of the clean speech signal of L-1 frame 1st frame of the voice signal of the continuous L frame entered is to the reverberation energy coefficient of the clean speech signal of L-1 frame closest to actual feelings Condition, is believed by the continuous L frame voice of its voice-input device 1 obtained, voice-input device 2 and voice-input device 3 input Number the voice signal of the energy of reverb signal that produces at L to L-1 frame clean speech signal of the 1st frame and actual L-th frame In the error of energy of reverb signal minimum.

If it should be noted that only removing the 1st to the of the continuous L frame voice signal of i-th voice-input device input The reverb signal that the clean speech signal of L-1 frame produces in L-th frame, then, in this step, it is only necessary to obtain i-th voice The difference of the energy of the continuous L frame signal between any two of input equipment and other M-1 voice-input device.

As a example by triple channel, if only removing the 1st frame of the continuous L frame voice signal of the 1st voice-input device input extremely The reverb signal that the clean speech signal of L-1 frame produces in L-th frame, then have only to be obtained by formula 4.4 and formula 4.6 e_1,2And e_3,1?.

404, according to described i-th voice-input device and other M-1 voice-input device continuous L frame between any two The difference of the energy of signal, it is thus achieved that three formula, described 3rd formula is:

Wherein, as 1 ＜=l ＜=L-1, describedPure for i-th voice-input device l frame The updated value of the reverberation energy coefficient of clean voice signal, described g_i,lFor i-th voice-input device l frame clean speech signal Reverberation energy coefficient, described μ_lThe renewal coefficient of the l frame for pre-seting, described h (e_i, E_l) it is according to i-th phonetic entry The difference of the energy of equipment and other M-1 voice-input device continuous L frame signal between any two, and other M-1 voice The value that 1st frame of the continuous L frame voice signal of input equipment to the energy of the voice signal of L-th frame obtains.

According to described 3rd formula, it is thus achieved that the 1st of the described continuous L frame voice signal of described i-th voice-input device Frame is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the updated value of the coefficient of the speech signal energy of L-th frame.

It should be noted that according to formula 4.2, for making e_i,jMinimum, except needs i-th voice-input device inputs The reverberation energy coefficient g of clean speech signal of 1 to L-1 frame of continuous L frame voice signal_{I, l}, and jth voice is defeated Enter the reverberation energy coefficient g of the clean speech signal of 1 to the L-1 frame of the continuous L frame voice signal of equipment input_{J, l}, wherein, 1 ＜=l ＜=L-1, in addition it is also necessary to the coefficient g of the L-th frame speech signal energy of i-th voice-input device input_{I, L}With jth language The coefficient g of the L-th frame speech signal energy of sound input equipment input_{J, L}。

Therefore, i-th voice-input device and the continuous L frame voice signal of jth voice-input device input are being carried out 1 to L-1 frame clean speech signal reverberation energy coefficient renewal process in, it is also desirable to i-th phonetic entry is set The coefficient g of the L-th frame speech signal energy of the continuous L frame voice signal that standby and jth voice-input device inputs_{I, L}And g_{J, L}Enter Row updates, concrete, g_{I, L}And g_{J, L}Initial value may be configured as 1.

As a example by the dual pathways, according to formula 4.3, work as e_1,2Absolute value minimum time, it is thus achieved that voice-input device 1 and language 1st frame of the voice signal of sound input equipment 2 input is closest real to the reverberation energy coefficient of the clean speech signal of L-1 frame Actual value, based on this, can be constructed as follows the function stating formula 4.7:

J_1,2=e_1,2 ² 4.7

Concrete, J can be taken_i,jFor e_{I, j}2n power, wherein, n ＞=1.

Pass through J_1,2Respectively to the 1st frame of the voice signal that voice-input device 1 and voice-input device 2 input to L-1 The coefficient of the reverberation energy coefficient of the clean speech signal of frame and the speech signal energy of L-th frame seeks local derviation, obtains formula 4.8 With formula 4.9:

$\frac{{\∂J}_{\mathrm{1,2}}}{{\∂g}_{1,l}}=-2*e_{\mathrm{1,2}}*E_{2,l},l=\mathrm{1,2},..L---4.8$

$\frac{{\∂J}_{\mathrm{1,2}}}{{\∂g}_{2,l}}=2*e_{\mathrm{1,2}}*E_{1,l},l=\mathrm{1,2},..L---4.9$

According to formula 4.8 and formula 4.9, it is readily obtained the 1st frame of voice-input device 1 and voice-input device 2 to The more new formula 4.10 of the coefficient of the reverberation energy coefficient of the clean speech energy of L-1 frame and the speech signal energy of L-th frame With formula 4.11:

${g_{1,l}}^{,\′}=g_{1,l}+{\mathrm{\μ}}_l*\frac{e_{\mathrm{1,2}}*E_{2,l}}{\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{E}_{2,l}*E_{2,l}},l=\mathrm{1,2},..L---4.10$

${g_{2,l}}^{,\′}=g_{2,l}+{\mathrm{\μ}}_l*\frac{e_{\mathrm{1,2}}*E_{1,l}}{\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{E}_{1,l}*E_{1,l}},l=\mathrm{1,2},...,L---4.11$

Wherein, as 1 ＜=l ＜=L-1, g_1,l' and g_2,l' respectively be update after voice-input device 1 and voice defeated Enter the reverberation energy coefficient of the clean speech signal of the l frame signal of the voice signal of equipment 2 input, g_1,lAnd g_2,lIt is more respectively Mixing of the clean speech signal of the l frame signal of the voice signal of new front voice-input device 1 and voice-input device 2 input Ring energy coefficient, μ_lIt is the renewal coefficient corresponding to l frame signal pre-seted, as l=L, g_1,l' and g_2,l' it is more respectively The speech signal energy of the L-th frame signal of the voice signal of voice-input device 1 after Xin and voice-input device 2 input be Number, g_1,lAnd g_2,lIt is the L-th frame letter of the voice signal of the voice-input device 1 before updating and voice-input device 2 input respectively Number the coefficient of speech signal energy, μ_lIt it is the renewal coefficient corresponding to l frame signal pre-seted.

It should be noted that combine formula 4.3, as the g of voice-input device 1_1,lG with voice-input device 2_2,lValue When being 0, the e obtained_1,2Absolute value minimum, be 0, but, this is the situation of a kind of mistake, for avoiding sending out of this situation Raw, to g_1,lAnd g_2,lAfter being updated, need g_1,lAnd g_2,lIt is normalized, concrete, to g_1,lAnd g_2,lCarry out Normalized is the known technology of those skilled in the art, and this is repeated no more by the embodiment of the present invention.

According to formula 4.10, for example, when the reverberation time is 16 frame, the 1st frame of voice-input device 1 input is arrived In the renewal process of the reverberation energy coefficient of the clean speech signal of the 15th frame, the parameter that needs are used is: the 1st frame is to the 15th frame The initial value of reverberation energy coefficient of clean speech signal, the renewal coefficient μ corresponding to L-th frame signal pre-seted_l, e_1,2 And E_2,l, wherein, e_1,2By initial value and the 16th frame voice signal energy of the reverberation energy coefficient of 1-15 frame clean speech signal The initial value of the coefficient of amount, and the energy acquisition of the voice signal of 1-16 frame.

During carrying out the 17th frame voice signal removal reverberation, according to formula 4.10, pure to the 16th frame to the 2nd frame During the reverberation energy coefficient of clean voice signal is updated, the parameter used is needed to be: pure to the 16th frame of the 2nd frame The initial value (being the 1st frame updated value to the clean speech signal of the 15th frame) of the reverberation energy coefficient of voice signal, presets The renewal coefficient μ corresponding to L-th frame signal put_l, e_1,2And E_2,l, wherein, e_1,2Reverberation by 2-16 frame clean speech signal Initial value (the renewal of the i.e. the 16th frame speech signal energy of the initial value of energy coefficient and the coefficient of the 17th frame speech signal energy Value), and the energy acquisition of the voice signal of 2-17 frame.

Can obtain by that analogy, carry out the process that 1 to L-1 frame clean speech signal reverberation energy coefficient updates In, it is also desirable to the coefficient of L-th frame speech signal energy is updated.

The most for example, as a example by triple channel, to voice-input device 1, voice-input device 2 and voice-input device 3 1st frame of the voice signal of input is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the voice signal of L-th frame The process that the coefficient of energy is updated is:

In conjunction with formula 4.4 to formula 4.6, as described e_1,2, described e_2,3With described e_3,1Absolute value minimum time, it is thus achieved that 1st frame of the voice signal of voice-input device 1, voice-input device 2 and voice-input device 3 input is pure to L-1 frame The reverberation energy coefficient of clean voice signal, closest to actual value, based on this, can be constructed as follows the letter stating formula 4.12-4.14 Number:

J_1,2=e_1,2 ² 4.12

J_2,3=e_2,3 ² 4.13

J_3,1=e_3,1 ² 4.14

Pass through J_1,2Respectively to the 1st frame of the voice signal that voice-input device 1 and voice-input device 2 input to L-1 The coefficient of the reverberation energy coefficient of the clean speech signal of frame and the speech signal energy of L-th frame seeks local derviation, passes through J_2,3Respectively To the clean speech signal of the 1st frame of the voice signal that voice-input device 2 and voice-input device 3 input to L-1 frame The coefficient of the speech signal energy of reverberation energy coefficient and L-th frame seeks local derviation, passes through J_3,1Respectively to voice-input device 3 He 1st frame of the voice signal of voice-input device 1 input is to the reverberation energy coefficient and the of the clean speech signal of L-1 frame The coefficient of the speech signal energy of L frame seeks local derviation, obtains formula 4.15-formula 4.20:

$\frac{{\∂J}_{\mathrm{1,2}}}{{\∂g}_{1,l}}=-2*e_{\mathrm{1,2}}*E_{2,l},l=\mathrm{1,2},..L---4.15$

$\frac{{\∂J}_{\mathrm{1,2}}}{{\∂g}_{2,l}}=2*e_{\mathrm{1,2}}*E_{1,l},l=\mathrm{1,2},..L---4.16$

$\frac{{\∂J}_{2,3}}{{\∂g}_{2,l}}=-2*e_{2,3}*E_{3,l},l=\mathrm{1,2},..L---4.17$

$\frac{{\∂J}_{\mathrm{2,3}}}{{\∂g}_{3,l}}=2*e_{2,3}*E_{2,l},l=\mathrm{1,2},..L---4.18$

$\frac{{\∂J}_{3,1}}{{\∂g}_{3,l}}=-2*e_{1,2}*E_{1,l},l=\mathrm{1,2},..L---4.19$

$\frac{{\∂J}_{3,1}}{{\∂g}_{1,l}}=2*e_{1,2}*E_{3,l},l=\mathrm{1,2},..L---4.20$

It is readily obtained voice-input device 1, voice-input device 2 and phonetic entry according to formula 4.15 to formula 4.20 to set 1st frame of standby 3 is to the more new formula 4.21 of the reverberation energy coefficient of L-th frame clean speech energy to formula 4.23:

$\begin{array}{c}{g}_{1,l}^{,}=g_{1,l}+{\mathrm{\μ}}_l*\frac{e_{\mathrm{1,2}}*E_{2,l}}{\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{E}_{2,l}*E_{2,l}}\\ -{\mathrm{\μ}}_l*\frac{e_{\mathrm{3,1}}*E_{3,l}}{\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{E}_{3,l}*E_{3,l}},l=\mathrm{1,2},..L\end{array}---4.21$

$\begin{array}{c}{g}_{2,l}^{,}=g_{2,l}\left(i\right)+{\mathrm{\μ}}_l*\frac{e_{\mathrm{1,2}}*E_{1,l}}{\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{E}_{1,l}*E_{1,l}}\\ +{\mathrm{\μ}}_l*\frac{e_{2,3}*E_{3,l}}{\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{E}_{3,l}*E_{3,l}},l=\mathrm{1,2},...,L\end{array}---4.22$

$\begin{array}{c}{g}_{3,l}^{,}=g_{3,l}+{\mathrm{\μ}}_l*\frac{e_{3,1}*E_{1,l}}{\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{E}_{1,l}*E_{1,l}}\\ -{\mathrm{\μ}}_l*\frac{e_{2,3}*E_{2,l}}{\underset{l=1}{\overset{L}{\mathrm{\Σ}}}{E}_{2,l}*E_{2,l}},l=\mathrm{1,2},..L\end{array}---4.23$

Wherein, as 1 ＜=l ＜=L-1, g_1,l’、g_2,l' andIt is the voice-input device 1 after updating, voice respectively The reverberation energy system of the clean speech signal of the l frame signal of the voice signal of input equipment 2 and voice-input device 3 input Number, g_1,l、g_2,lAnd g_3,lIt is the voice-input device 1 before updating, voice-input device 2 and voice-input device 3 input respectively The reverberation energy coefficient of the clean speech signal of the l frame signal of voice signal, as l=L, g_1,l’、g_2,l' andIt is respectively The L-th frame signal of the voice signal of voice-input device 1, voice-input device 2 and voice-input device 3 input after renewal The coefficient of speech signal energy, g_1,l、g_2,lAnd g_3,lIt is voice-input device 1, voice-input device 2 and the language before updating respectively The coefficient of the speech signal energy of the L-th frame signal of the voice signal of sound input equipment 3 input, μ_lIt it is the l frame letter pre-seted Number corresponding renewal coefficient.

Same, to the g after updating_1,l’、g_2,l' andNormalized to be done, the embodiment of the present invention is to the most superfluous State.

It should be noted that either utilize dual pathways least mean square algorithm or triple channel least mean square algorithm, one In individual fixing environment, such as in a fixing conferencing environment, it is contemplated that from L-th frame voice signal more close to frame, such as L-1 frame, L-2 frame, its reverb signal produced in L-th frame change to the reverb signal that the voice signal of L-th frame produces Relatively big, from L-th frame voice signal frame farther out, such as the 1st frame or the 2nd frame, the change of its reverb signal produced in L-th frame is relatively Little, and the direct sound wave that sound source inputs in L-th frame, i.e. the changing greatly of clean speech signal of L-th frame input, based on this, Pre-set renewal coefficient μ_lTime, μ can be made_L≥μ_L-1≥μ_L-2≥...≥μ₁.Concrete, can be according to practical situation at reasonable model Enclose interior to μ_lCarrying out value, this is not limited by the embodiment of the present invention.By this method, L-th frame can quickly be followed the tracks of The change of the reverb signal that the clean speech of direct sound wave and the frame close to L-th frame produces in L-th frame, and can ensure that from The stability of the reverb signal that the clean speech signal of L-th frame frame farther out produces in L-th frame.

405, the 1st frame of the continuous L frame voice signal inputted according to described i-th voice-input device is to L-1 frame The 1st frame after the energy of clean speech signal, and the renewal of described i-th voice-input device is to the clean speech of L-1 frame The reverberation energy coefficient of signal, estimates that the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs is to The clean speech signal of the L-1 frame reverberation energy to the reverb signal that described L-th frame voice signal produces.

406, according to described i-th voice-input device input continuous L frame voice signal L-th frame signal energy and Described reverberation energy, it is thus achieved that the energy of the clean speech signal of described L-th frame signal.

Concrete, the detailed implementation process of step 406 and step 407 can be found in the step 303 in above-described embodiment and step Rapid 304, this is repeated no more by the embodiment of the present invention.

Embodiments provide a kind of method eliminating reverberation, based on the method, pure to the 1st frame to L-1 frame The reverberation energy coefficient of clean voice signal has carried out adaptive updates, the clean speech signal pair of the 1st frame obtained to L-1 frame The reverberation energy of the reverb signal that described L-th frame signal produces is more accurate, therefore, with it, can be more accurately The energy of the reverb signal in acquisition voice signal, and eliminate the impact of this reverb signal, thus improve voice quality.

Embodiments providing a kind of method eliminating reverberation, in conjunction with Fig. 5, the method includes:

501, the 1st frame of continuous L frame voice signal of i-th and other M-1 voice-input device input is obtained extremely The amplitude of the voice signal of L-th frame, and the continuous L frame voice letter of described i-th and other M-1 voice-input device input Number the 1st frame to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and L-th frame clean speech signal amplitude be Number, and the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to the clean speech signal of L-1 frame Amplitude, wherein, described voice signal comprises clean speech signal and reverb signal, and described continuous L frame voice signal is frequency domain letter Number, described L is more than 1.

502, according to described i-th and the 1st of the continuous L frame voice signal of other M-1 voice-input device input the Frame is to the amplitude of the voice signal of L-th frame, and described i-th and the continuous L frame language of other M-1 voice-input device input 1st frame of the tone signal reverberation amplitude coefficient to the clean speech signal of L-1 frame and the clean speech signal amplitude of L-th frame Coefficient, to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to the clean speech of L-1 frame The coefficient of the reverberation amplitude coefficient of signal and the clean speech signal amplitude of L-th frame is updated.

503, the 1st frame of the continuous L frame voice signal inputted according to described i-th voice-input device is to L-1 frame The 1st frame after the amplitude of clean speech signal, and the renewal of described i-th voice-input device is to the clean speech of L-1 frame The reverberation amplitude coefficient of signal, estimates that the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs is to The clean speech signal of the L-1 frame reverberation amplitude to the reverb signal that described L-th frame voice signal produces.

504, according to described i-th voice-input device input continuous L frame voice signal L-th frame signal amplitude and Described reverberation amplitude, it is thus achieved that the amplitude of the clean speech signal of described L-th frame signal.

In the above-described embodiments, from the angle of energy, the voice signal of the L-th frame of continuous L frame voice signal is carried out reverberation Eliminate, optionally, it is also possible to from the angle of amplitude, the voice signal of the L-th frame of described continuous L frame voice signal is carried out reverberation Eliminating, concrete, the technical characteristic of step 501-504 can be found in the technology in the step 301-step 304 in above-described embodiment Feature, this is repeated no more by the embodiment of the present invention.

Embodiments providing a kind of method eliminating reverberation, the method includes: obtain i-th and other M-1 1st frame of the continuous L frame voice signal of individual voice-input device input is to the amplitude of the voice signal of L-th frame, and described i-th And the 1st frame of the continuous L frame voice signal of other M-1 voice-input device input is to the clean speech signal of L-1 frame Reverberation amplitude coefficient and the coefficient of clean speech signal amplitude of L-th frame, and the company of i-th voice-input device input Continuing the 1st frame amplitude to the clean speech signal of L-1 frame of L frame voice signal, wherein, described voice signal comprises pure language Tone signal and reverb signal, described continuous L frame voice signal is frequency-region signal, and described L is more than 1；According to described i-th and its 1st frame of the continuous L frame voice signal of his M-1 voice-input device input is to the amplitude of the voice signal of L-th frame and described 1st frame of the continuous L frame voice signal of i-th and the input of other M-1 voice-input device is to the clean speech of L-1 frame The reverberation amplitude coefficient of signal and the coefficient of the clean speech signal amplitude of L-th frame, defeated to described i-th voice-input device 1st frame of the continuous L frame voice signal entered is pure to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and L-th frame The coefficient of clean voice signal amplitude is updated；Continuous L frame voice signal according to the input of described i-th voice-input device 1st frame is to the amplitude of the clean speech signal of L-1 frame, and the 1st frame after the renewal of described i-th voice-input device is to The reverberation amplitude coefficient of the clean speech signal of L-1 frame, estimates the continuous L frame voice that described i-th voice-input device inputs 1st frame of signal is to reverberation width to the reverb signal that described L-th frame voice signal produces of the clean speech signal of L-1 frame Degree；The amplitude of the L-th frame signal of the continuous L frame voice signal according to the input of described i-th voice-input device and described reverberation Amplitude, it is thus achieved that the amplitude of the clean speech signal of described L-th frame signal.Based on the method, pure to the 1st frame to L-1 frame The reverberation amplitude coefficient of voice signal is updated, and the clean speech signal of the 1st frame obtained to L-1 frame is to described L The reverberation amplitude of the reverb signal that frame signal produces is more accurate, therefore, with it, voice can be obtained more accurately The amplitude of the reverb signal in signal, and eliminate the impact of this reverb signal, thus improve voice quality.

Embodiments providing a kind of device eliminating reverberation, in conjunction with Fig. 6, this device includes: first obtains unit 601, the first updating block 602, the first evaluation unit 603 and second obtains unit 604；

Described first obtains unit 601, for obtaining i-th and the continuous L of other M-1 voice-input device input 1st frame of frame voice signal is to the energy of the voice signal of L-th frame, and described i-th and other M-1 voice-input device 1st frame of the continuous L frame voice signal of input is to the reverberation energy coefficient of the clean speech signal of L-1 frame and L-th frame The coefficient of speech signal energy, and the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to L-1 frame The energy of clean speech signal, wherein, described voice signal comprises clean speech signal and reverb signal, described continuous L frame Voice signal is frequency-region signal, and described L is more than 1；

Described first updating block 602, for according to described first obtain unit 601 send described i-th and its 1st frame of the continuous L frame voice signal of his M-1 voice-input device input is to the energy of the voice signal of L-th frame and described 1st frame of the continuous L frame voice signal of i-th and the input of other M-1 voice-input device is to the clean speech of L-1 frame The reverberation energy coefficient of signal and the coefficient of the speech signal energy of L-th frame, to the input of described i-th voice-input device 1st frame of L frame voice signal is believed to the reverberation energy coefficient of the clean speech signal of L-1 frame and the voice of L-th frame continuously The coefficient of number energy is updated；

Described first evaluation unit 603, defeated for obtaining the described i-th voice of unit 601 transmission according to described first Enter the 1st frame of continuous L frame voice signal of equipment input to the energy of the clean speech signal of L-1 frame, and described first more The 1st frame after the renewal of the described i-th voice-input device that new unit 602 sends is to the clean speech signal of L-1 frame Reverberation energy coefficient, estimates that the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs is to L-1 frame The clean speech signal reverberation energy to the reverb signal that described L-th frame voice signal produces；

Described second obtains unit 604, defeated for obtaining the described i-th voice of unit 601 transmission according to described first Enter the energy of the L-th frame signal of the continuous L frame voice signal of equipment input and the described mixed of described first evaluation unit 603 transmission Ring energy, it is thus achieved that the energy of the clean speech signal of described L-th frame signal.

Optionally, described first obtain unit 601 specifically for: according to the Energy Decay Curve of default impulse signal Slope q, by the first formula, it is thus achieved that described i and the continuous L frame voice signal of other M-1 voice-input device input The 1st frame to the reverberation energy coefficient g of the clean speech signal of L-1 frame₁、g₂…g_L-1And the speech signal energy of L-th frame Coefficient g_L, wherein, described first formula is:

Wherein, 1 ＜=l ＜=L.

Concrete, described L is reverberation time T and the ratio of the predetermined time period of frame rounds, wherein, and the described reverberation time The slope q of the Energy Decay Curve of described default impulse signal according to T, the energy attenuation of the described impulse signal of estimation is extremely The time of preset energy.

Concrete, described first updating block 602 specifically for: obtain i-th voice-input device according to the second formula With the difference of the energy of the continuous L frame signal of jth voice-input device, wherein, described second formula is:

Wherein, as 1 ＜=l ＜=L-1, described g_j,lE_i,lDefeated for jth voice Enter the reverberation energy coefficient of the clean speech signal of the l frame of equipment and the signal energy of the l frame of i-th voice-input device Convolution, described g_i,lE_j,lThe reverberation energy coefficient of clean speech signal of l frame and jth for i-th voice-input device The convolution of the signal energy of the l frame of individual voice-input device, as l=L, described g_j,lE_i,lFor jth voice-input device The convolution of signal energy of L-th frame of coefficient and i-th voice-input device of speech signal energy of L-th frame, described g_{i, l}E_j,lThe L-th frame of coefficient and jth voice-input device of speech signal energy for the L-th frame of i-th voice-input device The convolution of signal energy；

Described i-th voice-input device is obtained with other M-1 voice-input device two-by-two according to described second formula Between the difference of energy of continuous L frame signal；

According to described i-th voice-input device and other M-1 voice-input device continuous L frame signal between any two The difference of energy, it is thus achieved that three formula, described 3rd formula is:

Wherein, as 1 ＜=l ＜=L-1, describedPure for i-th voice-input device l frame The updated value of the reverberation energy coefficient of clean voice signal, described g_i,lFor i-th voice-input device l frame clean speech signal Reverberation energy coefficient, described μ_lThe renewal coefficient of the l frame for pre-seting, described h (e_i, E_l) it is according to i-th phonetic entry The difference of the energy of equipment and other M-1 voice-input device continuous L frame signal between any two, and other M-1 voice The value that 1st frame of the continuous L frame voice signal of input equipment to the energy of the voice signal of L-th frame obtains；

According to described 3rd formula, it is thus achieved that the 1st of the described continuous L frame voice signal of described i-th voice-input device Frame is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the updated value of the coefficient of the speech signal energy of L-th frame.

Concrete, described μ_l＞=μ_l-1, wherein, described μ_lThe renewal coefficient of the l frame for pre-seting, described μ_l-1For in advance The renewal coefficient of the l-1 frame arranged, described l-1 frame is the former frame of l frame, described 1 ＜=l-1.

Embodiments providing a kind of device eliminating reverberation, this device includes: first obtains unit, and first updates Unit, the first evaluation unit and second obtains unit.Described first acquisition unit obtains i-th and other M-1 voice is defeated Enter the 1st frame of continuous L frame voice signal of equipment input to the energy of the voice signal of L-th frame, and described i-th and other 1st frame of the continuous L frame voice signal of M-1 voice-input device input is to the reverberation energy of the clean speech signal of L-1 frame The coefficient of the speech signal energy of coefficient of discharge and L-th frame, and the continuous L frame voice letter of i-th voice-input device input Number the 1st frame to the energy of the clean speech signal of L-1 frame, wherein, described voice signal comprises clean speech signal and mixed Ringing signal, described continuous L frame voice signal is frequency-region signal, and described L is more than 1；Described first updating block is according to described first Obtain the described i-th of unit transmission and the 1st frame of the continuous L frame voice signal of other M-1 voice-input device input To the energy of the voice signal of L-th frame, and the continuous L frame voice that described i-th and other M-1 voice-input device input 1st frame of signal to the reverberation energy coefficient of the clean speech signal of L-1 frame and L-th frame speech signal energy be Number, to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to the clean speech signal of L-1 frame Reverberation energy coefficient and the coefficient of speech signal energy of L-th frame be updated；Described first evaluation unit is according to described 1st frame of the first continuous L frame voice signal obtaining the described i-th voice-input device input that unit sends is to L-1 frame The energy of clean speech signal, and after the renewal of described i-th voice-input device that sends of described first updating block 1st frame, to the reverberation energy coefficient of the clean speech signal of L-1 frame, estimates the company that described i-th voice-input device inputs The reverb signal that described L-th frame voice signal is produced by the 1st frame of continuous L frame voice signal to the clean speech signal of L-1 frame Reverberation energy；The described i-th voice-input device that described second acquisition unit sends according to described first acquisition unit is defeated The described reverberation energy that the energy of the L-th frame signal of the continuous L frame voice signal entered and described first evaluation unit send, it is thus achieved that The energy of the clean speech signal of described L-th frame signal.Based on this device, to the clean speech signal of the 1st frame to L-1 frame Reverberation energy coefficient is updated, and described L-th frame signal is produced by the clean speech signal of the 1st frame obtained to L-1 frame The reverberation energy of reverb signal more accurate, therefore, by this device, can obtain in voice signal more accurately is mixed The energy of sound signal, and eliminate the impact of this reverb signal, thus improve voice quality.

Embodiments providing a kind of device eliminating reverberation, in conjunction with Fig. 7, this device includes: the 3rd obtains unit 701, the second updating block 702, the second evaluation unit 703 and the 4th acquisition unit 704:

Described 3rd obtains unit 701, for obtaining i-th and the continuous L of other M-1 voice-input device input 1st frame of frame voice signal is to the amplitude of the voice signal of L-th frame, and described i-th and other M-1 voice-input device 1st frame of the continuous L frame voice signal of input is to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and L-th frame The coefficient of clean speech signal amplitude, and the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to the The amplitude of the clean speech signal of L-1 frame, wherein, described voice signal comprises clean speech signal and reverb signal, described company Continuous L frame voice signal is frequency-region signal, and described L is more than 1；

Described second updating block 702, for according to described 3rd obtain unit 701 send described i-th and its 1st frame of the continuous L frame voice signal of his M-1 voice-input device input is to the amplitude of the voice signal of L-th frame and described 1st frame of the continuous L frame voice signal of i-th and the input of other M-1 voice-input device is to the clean speech of L-1 frame The reverberation amplitude coefficient of signal and the coefficient of the clean speech signal amplitude of L-th frame, defeated to described i-th voice-input device 1st frame of the continuous L frame voice signal entered is pure to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and L-th frame The coefficient of clean voice signal amplitude is updated；

Described second evaluation unit 703, defeated for obtaining the described i-th voice of unit 701 transmission according to the described 3rd Enter the 1st frame of continuous L frame voice signal of equipment input to the amplitude of the clean speech signal of L-1 frame, and described second more The 1st frame after the renewal of the described i-th voice-input device that new unit 702 sends is to the clean speech signal of L-1 frame Reverberation amplitude coefficient, estimates that the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs is to L-1 frame The clean speech signal reverberation amplitude to the reverb signal that described L-th frame voice signal produces；

Described 4th obtains unit 704, defeated for obtaining the described i-th voice of unit 701 transmission according to the described 3rd Enter the amplitude of the L-th frame signal of the continuous L frame voice signal of equipment input and the described mixed of described second evaluation unit 703 transmission The amplitude of sound, it is thus achieved that the amplitude of the clean speech signal of described L-th frame signal.

Embodiments providing a kind of device eliminating reverberation, this device includes: the 3rd obtains unit, and second updates Unit, the second evaluation unit and the 4th obtains unit.Described 3rd acquisition unit obtains i-th and other M-1 voice is defeated Enter the 1st frame of continuous L frame voice signal of equipment input to the amplitude of the voice signal of L-th frame, and described i-th and other 1st frame of the continuous L frame voice signal of M-1 voice-input device input is to the reverberation width of the clean speech signal of L-1 frame The coefficient of the clean speech signal amplitude of degree coefficient and L-th frame, and the continuous L frame language of i-th voice-input device input 1st frame of tone signal is to the amplitude of the clean speech signal of L-1 frame, and wherein, described voice signal comprises clean speech signal And reverb signal, described continuous L frame voice signal is frequency-region signal, and described L is more than 1；Described second updating block is according to described 3rd obtains the of described i-th that unit sends and the continuous L frame voice signal that other M-1 voice-input device inputs 1 frame is to the amplitude of the voice signal of L-th frame, and described i-th and the continuous L frame of other M-1 voice-input device input 1st frame of the voice signal reverberation amplitude coefficient to the clean speech signal of L-1 frame and the clean speech signal width of L-th frame The coefficient of degree, to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to the pure language of L-1 frame The coefficient of the reverberation amplitude coefficient of tone signal and the clean speech signal amplitude of L-th frame is updated；Described second pro form bill Unit is according to the 1st frame of the described 3rd continuous L frame voice signal obtaining the described i-th voice-input device input that unit sends To the amplitude of the clean speech signal of L-1 frame, and the described i-th voice-input device that sends of described second updating block The 1st frame after renewal, to the reverberation amplitude coefficient of the clean speech signal of L-1 frame, estimates described i-th voice-input device Described L-th frame voice signal is produced by the 1st frame of the continuous L frame voice signal of input to the clean speech signal of L-1 frame The reverberation amplitude of reverb signal；The described i-th voice that described 4th acquisition unit sends according to described 3rd acquisition unit is defeated Enter amplitude and the described reverberation of described second evaluation unit transmission of the L-th frame signal of the continuous L frame voice signal of equipment input Amplitude, it is thus achieved that the amplitude of the clean speech signal of described L-th frame signal.Based on this device, pure to the 1st frame to L-1 frame The reverberation amplitude coefficient of voice signal is updated, and the clean speech signal of the 1st frame obtained to L-1 frame is to described L The reverberation amplitude of the reverb signal that frame signal produces is more accurate, therefore, by this device, can obtain voice more accurately The amplitude of the reverb signal in signal, and eliminate the impact of this reverb signal, thus improve voice quality.

Embodiments providing a kind of equipment eliminating reverberation, in conjunction with Fig. 8, this equipment includes: first transceiver 801 With first processor 802；

Described first transceiver 801 is used for: receive i-th voice-input device and other M-1 voice-input device The continuous L frame voice signal of input；

Described first processor 802 is used for: obtain i-th and the continuous L frame of other M-1 voice-input device input 1st frame of voice signal to the energy of the voice signal of L-th frame, and described i-th and other M-1 voice-input device defeated 1st frame of the continuous L frame voice signal entered is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the language of L-th frame The coefficient of tone signal energy, and the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to L-1 frame The energy of clean speech signal, wherein, described voice signal comprises clean speech signal and reverb signal, described continuous L frame language Tone signal is frequency-region signal, and described L is more than 1；

The 1st frame according to described i-th and the continuous L frame voice signal of other M-1 voice-input device input is extremely The energy of the voice signal of L-th frame, and the continuous L frame voice letter of described i-th and other M-1 voice-input device input Number the 1st frame to the reverberation energy coefficient of the clean speech signal of L-1 frame and the coefficient of the speech signal energy of L-th frame, Clean speech signal to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to L-1 frame The coefficient of the speech signal energy of reverberation energy coefficient and L-th frame is updated；

1st frame of the continuous L frame voice signal according to the input of described i-th voice-input device is pure to L-1 frame The 1st frame after the energy of voice signal, and the renewal of described i-th voice-input device is to the clean speech signal of L-1 frame Reverberation energy coefficient, estimate that the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs is to L-1 frame The clean speech signal reverberation energy to the reverb signal that described L-th frame voice signal produces；

The energy of L-th frame signal of the continuous L frame voice signal according to the input of described i-th voice-input device and described Reverberation energy, it is thus achieved that the energy of the clean speech signal of described L-th frame signal.

Optionally, described first processor 802 is additionally operable to:

The slope q of the Energy Decay Curve according to default impulse signal, by the first formula, it is thus achieved that described i and 1st frame of the continuous L frame voice signal of other M-1 voice-input device inputs is to clean speech signal mixed of L-1 frame Ring energy coefficient g₁、g₂…g_L-1And the coefficient g of the speech signal energy of L-th frame_L, wherein, described first formula is:

Wherein, 1 ＜=l ＜=L.

Concrete, described L is reverberation time T and the ratio of the predetermined time period of frame rounds, wherein, and the described reverberation time The slope q of the Energy Decay Curve of described default impulse signal according to T, the energy attenuation of the described impulse signal of estimation is extremely The time of preset energy.

Concrete, described first processor 802 is used for:

The energy of the continuous L frame signal of i-th voice-input device and jth voice-input device is obtained according to the second formula The difference of amount, wherein, described second formula is:

Wherein, as 1 ＜=l ＜=L-1, described g_j,lE_i,lDefeated for jth voice Enter the reverberation energy coefficient of the clean speech signal of the l frame of equipment and the signal energy of the l frame of i-th voice-input device Convolution, described g_i,lE_j,lThe reverberation energy coefficient of clean speech signal of l frame and jth for i-th voice-input device The convolution of the signal energy of the l frame of individual voice-input device, as l=L, described g_j,lE_i,lFor jth voice-input device The convolution of signal energy of L-th frame of coefficient and i-th voice-input device of speech signal energy of L-th frame, described g_{i, l}E_j,lThe L-th frame of coefficient and jth voice-input device of speech signal energy for the L-th frame of i-th voice-input device The convolution of signal energy；

Described i-th voice-input device is obtained with other M-1 voice-input device two-by-two according to described second formula Between the difference of energy of continuous L frame signal；

According to described i-th voice-input device and other M-1 voice-input device continuous L frame signal between any two The difference of energy, it is thus achieved that three formula, described 3rd formula is:

Wherein, as 1 ＜=l ＜=L-1, describedPure for i-th voice-input device l frame The updated value of the reverberation energy coefficient of clean voice signal, described g_i,lFor i-th voice-input device l frame clean speech signal Reverberation energy coefficient, described μ_lThe renewal coefficient of the l frame for pre-seting, described h (e_i, E_l) it is according to i-th phonetic entry The difference of the energy of equipment and other M-1 voice-input device continuous L frame signal between any two, and other M-1 voice The value that 1st frame of the continuous L frame voice signal of input equipment to the energy of the voice signal of L-th frame obtains；

According to described 3rd formula, it is thus achieved that the 1st of the described continuous L frame voice signal of described i-th voice-input device Frame is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the updated value of the coefficient of the speech signal energy of L-th frame.

Concrete, described μ_l＞=μ_l-1, wherein, described μ_lThe renewal coefficient of the l frame for pre-seting, described μ_l-1For in advance The renewal coefficient of the l-1 frame arranged, described l-1 frame is the former frame of l frame, described 1 ＜=l-1.

Embodiments providing a kind of equipment eliminating reverberation, this equipment includes: first transceiver and first processes Device.Described first transceiver receives i-th voice-input device and the continuous L frame of other M-1 voice-input device input Voice signal；Described first processor obtains the continuous L frame voice letter of i-th and other M-1 voice-input device input Number the 1st frame to the energy of the voice signal of L-th frame, and described i-th and the company of other M-1 voice-input device input 1st frame of continuous L frame voice signal is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the voice signal of L-th frame The coefficient of energy, and the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to the pure language of L-1 frame The energy of tone signal, wherein, described voice signal comprises clean speech signal and reverb signal, described continuous L frame voice signal For frequency-region signal, described L is more than 1；According to described i-th and the continuous L frame voice of other M-1 voice-input device input 1st frame of signal is to the energy of the voice signal of L-th frame, and described i-th and other M-1 voice-input device input 1st frame of L frame voice signal is believed to the reverberation energy coefficient of the clean speech signal of L-1 frame and the voice of L-th frame continuously The coefficient of number energy, pure to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to L-1 frame The coefficient of the reverberation energy coefficient of clean voice signal and the speech signal energy of L-th frame is updated；According to described i-th language 1st frame of the continuous L frame voice signal of sound input equipment input to the energy of the clean speech signal of L-1 frame, and described i-th The 1st frame after the renewal of individual voice-input device, to the reverberation energy coefficient of the clean speech signal of L-1 frame, estimates described 1st frame of the continuous L frame voice signal of i voice-input device input to the clean speech signal of L-1 frame to described L-th frame The reverberation energy of the reverb signal that voice signal produces；Continuous L frame voice letter according to the input of described i-th voice-input device Number the energy of L-th frame signal and described reverberation energy, it is thus achieved that the energy of the clean speech signal of described L-th frame signal.Based on This equipment, is updated the reverberation energy coefficient of the 1st frame to the clean speech signal of L-1 frame, the 1st frame obtained to The clean speech signal of L-1 frame is more accurate to the reverberation energy of the reverb signal that described L-th frame signal produces, and therefore, passes through This equipment, can obtain the energy of reverb signal in voice signal more accurately, and eliminates the impact of this reverb signal, from And improve voice quality.

Embodiments providing a kind of equipment eliminating reverberation, in conjunction with Fig. 9, this equipment includes: second transceiver 901 With the second processor 902；

Described second transceiver 901 is used for: receive i-th voice-input device and other M-1 voice-input device The continuous L frame voice signal of input；

Described second processor 902 is used for: obtain i-th and the continuous L frame of other M-1 voice-input device input 1st frame of voice signal to the amplitude of the voice signal of L-th frame, and described i-th and other M-1 voice-input device defeated 1st frame of the continuous L frame voice signal entered is pure to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and L-th frame The coefficient of clean voice signal amplitude, and the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to L-1 The amplitude of the clean speech signal of frame, wherein, described voice signal comprises clean speech signal and reverb signal, described continuous L Frame voice signal is frequency-region signal, and described L is more than 1；

The 1st frame according to described i-th and the continuous L frame voice signal of other M-1 voice-input device input is extremely The amplitude of the voice signal of L-th frame, and the continuous L frame voice letter of described i-th and other M-1 voice-input device input Number the 1st frame to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and L-th frame clean speech signal amplitude be Number, to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to the clean speech signal of L-1 frame Reverberation amplitude coefficient and the coefficient of clean speech signal amplitude of L-th frame be updated；

1st frame of the continuous L frame voice signal according to the input of described i-th voice-input device is pure to L-1 frame The 1st frame after the amplitude of voice signal, and the renewal of described i-th voice-input device is to the clean speech signal of L-1 frame Reverberation amplitude coefficient, estimate that the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs is to L-1 frame The clean speech signal reverberation amplitude to the reverb signal that described L-th frame voice signal produces；

The amplitude of L-th frame signal of the continuous L frame voice signal according to the input of described i-th voice-input device and described Reverberation amplitude, it is thus achieved that the amplitude of the clean speech signal of described L-th frame signal.

Embodiments providing a kind of equipment eliminating reverberation, this equipment includes: second transceiver and second processes Device.Described second transceiver receives i-th voice-input device and the continuous L frame of other M-1 voice-input device input Voice signal；Described second processor obtains the continuous L frame voice letter of i-th and other M-1 voice-input device input Number the 1st frame to the amplitude of the voice signal of L-th frame, and described i-th and the company of other M-1 voice-input device input 1st frame of continuous L frame voice signal is to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and the clean speech of L-th frame The coefficient of signal amplitude, and pure to L-1 frame of the 1st frame of the continuous L frame voice signal of i-th voice-input device input The amplitude of clean voice signal, wherein, described voice signal comprises clean speech signal and reverb signal, described continuous L frame voice Signal is frequency-region signal, and described L is more than 1；According to described i-th and the continuous L frame of other M-1 voice-input device input 1st frame of voice signal to the amplitude of the voice signal of L-th frame, and described i-th and other M-1 voice-input device defeated 1st frame of the continuous L frame voice signal entered is pure to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and L-th frame The coefficient of clean voice signal amplitude, to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to the The coefficient of the reverberation amplitude coefficient of the clean speech signal of L-1 frame and the clean speech signal amplitude of L-th frame is updated；Root According to the 1st frame of continuous L frame voice signal of described i-th voice-input device input to the clean speech signal of L-1 frame The 1st frame after amplitude, and the renewal of described i-th voice-input device is to the reverberation amplitude of the clean speech signal of L-1 frame Coefficient, estimates the 1st frame clean speech to L-1 frame of the continuous L frame voice signal that described i-th voice-input device inputs The signal reverberation amplitude to the reverb signal that described L-th frame voice signal produces；Input according to described i-th voice-input device The amplitude of L-th frame signal of continuous L frame voice signal and described reverberation amplitude, it is thus achieved that the clean speech of described L-th frame signal The amplitude of signal.Based on this equipment, the reverberation amplitude coefficient of the 1st frame to the clean speech signal of L-1 frame is updated, The 1st frame obtained to L-1 frame clean speech signal to the reverberation amplitude of the reverb signal that described L-th frame signal produces more Accurately, therefore, by this equipment, the amplitude of reverb signal in voice signal can be obtained more accurately, and eliminate this and mix Ring the impact of signal, thus improve voice quality.

One of ordinary skill in the art will appreciate that: all or part of step realizing said method embodiment can be passed through The hardware that programmed instruction is relevant completes, and aforesaid program can be stored in a computer read/write memory medium, this program Upon execution, perform to include the step of said method embodiment；And aforesaid storage medium includes: ROM, RAM, magnetic disc or light The various medium that can store program code such as dish.

The above, the only detailed description of the invention of the present invention, but protection scope of the present invention is not limited thereto, and any Those familiar with the art, in the technical scope that the invention discloses, can readily occur in change or replace, should contain Cover within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with described scope of the claims.

Claims (18)

1. the method eliminating reverberation, it is characterised in that the method includes:

Obtain i-th and the 1st frame of continuous L frame voice signal that other M-1 voice-input device inputs is to the language of L-th frame The energy of tone signal, and the 1st frame of the continuous L frame voice signal of described i-th and other M-1 voice-input device input To the reverberation energy coefficient of clean speech signal of L-1 frame and the coefficient of the speech signal energy of L-th frame, and i-th 1st frame of the continuous L frame voice signal of voice-input device input is to the energy of the clean speech signal of L-1 frame, wherein, institute Predicate tone signal comprises clean speech signal and reverb signal, and described continuous L frame voice signal is frequency-region signal, and described L is more than 1；

The 1st frame according to described i-th and the continuous L frame voice signal of other M-1 voice-input device input is to L-th frame The energy of voice signal, and the continuous L frame voice signal of described i-th and other M-1 voice-input device input 1st frame is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the coefficient of the speech signal energy of L-th frame, to institute State the 1st frame reverberation to the clean speech signal of L-1 frame of the continuous L frame voice signal that i-th voice-input device inputs The coefficient of the speech signal energy of energy coefficient and L-th frame is updated；

1st frame of the continuous L frame voice signal according to the input of described i-th voice-input device is to the clean speech of L-1 frame The 1st frame after the energy of signal, and the renewal of described i-th voice-input device is to clean speech signal mixed of L-1 frame Ring energy coefficient, estimate pure to L-1 frame of the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs The clean voice signal reverberation energy to the reverb signal that described L-th frame voice signal produces；

The energy of the L-th frame signal of the continuous L frame voice signal according to the input of described i-th voice-input device and described reverberation Energy, it is thus achieved that the energy of the clean speech signal of described L-th frame signal.

Method the most according to claim 1, it is characterised in that described acquisition i-th and other M-1 phonetic entry set 1st frame of the continuous L frame voice signal of standby input is to the reverberation energy coefficient of the clean speech signal of L-1 frame and L-th frame The coefficient of speech signal energy specifically include:

The slope q of the Energy Decay Curve according to default impulse signal, by the first formula, it is thus achieved that described i-th and its 1st frame of the continuous L frame voice signal of his M-1 voice-input device input is to the reverberation of the clean speech signal of L-1 frame Energy coefficient g₁、g₂…g_L-1And the coefficient g of the speech signal energy of L-th frame_L, wherein, described first formula is:

Wherein, 1 ＜=l ＜=L.

Method the most according to claim 1 and 2, it is characterised in that described L is that reverberation time T is long with the Preset Time of frame The ratio of degree rounds, wherein, and the slope q of the Energy Decay Curve of described default impulse signal according to described reverberation time T, The energy attenuation of the described impulse signal estimated is to time of preset energy.

Method the most according to claim 1 and 2, it is characterised in that input described i-th voice-input device is continuous 1st frame of L frame voice signal is updated specifically including to the reverberation energy coefficient of the clean speech signal of L-1 frame:

The energy of the continuous L frame signal of i-th voice-input device and jth voice-input device is obtained according to the second formula Difference, wherein, described second formula is:

Wherein, as 1 ＜=l ＜=L-1, described g_j,lE_i,lSet for jth phonetic entry The volume of the signal energy of the reverberation energy coefficient of the clean speech signal of standby l frame and the l frame of i-th voice-input device Long-pending, described g_i,lE_j,lReverberation energy coefficient and the jth language of clean speech signal for the l frame of i-th voice-input device The convolution of the signal energy of the l frame of sound input equipment, as l=L, described g_j,lE_i,lL for jth voice-input device The convolution of the signal energy of the L-th frame of the coefficient of the speech signal energy of frame and i-th voice-input device, described g_i,lE_j,lFor The signal of the L-th frame of the coefficient of the speech signal energy of the L-th frame of i-th voice-input device and jth voice-input device The convolution of energy；

Described i-th voice-input device is obtained with other M-1 voice-input device between any two according to described second formula The difference of energy of continuous L frame signal；

Energy according to described i-th voice-input device He other M-1 voice-input device continuous L frame signal between any two The difference of amount, it is thus achieved that the 3rd formula, described 3rd formula is:

g’_i,l=g_i,l±μ_l* h, wherein, as 1 ＜=l ＜=L-1, described g '_i,lPure for i-th voice-input device l frame The updated value of the reverberation energy coefficient of clean voice signal, described g_i,lFor i-th voice-input device l frame clean speech signal Reverberation energy coefficient, described μ_lThe renewal coefficient of the l frame for pre-seting, described h (e_i, E_l) it is according to i-th phonetic entry The difference of the energy of equipment and other M-1 voice-input device continuous L frame signal between any two, and other M-1 voice The value that 1st frame of the continuous L frame voice signal of input equipment to the energy of the voice signal of L-th frame obtains；

According to described 3rd formula, it is thus achieved that the 1st frame of the described continuous L frame voice signal of described i-th voice-input device is extremely The updated value of the coefficient of the reverberation energy coefficient of the clean speech signal of L-1 frame and the speech signal energy of L-th frame.

Method the most according to claim 4, it is characterised in that described μ_l＞=μ_l-1, wherein, described μ_lFor pre-set The renewal coefficient of l frame, described μ_l-1The renewal coefficient of the l-1 frame for pre-seting, described l-1 frame is the former frame of l frame, Described 1 ＜=l-1.

6. the method eliminating reverberation, it is characterised in that the method includes:

Obtain i-th and the 1st frame of continuous L frame voice signal that other M-1 voice-input device inputs is to the language of L-th frame The amplitude of tone signal, and the 1st frame of the continuous L frame voice signal of described i-th and other M-1 voice-input device input To the reverberation amplitude coefficient of clean speech signal of L-1 frame and the coefficient of the clean speech signal amplitude of L-th frame, Yi Ji 1st frame of the continuous L frame voice signal of i voice-input device input to the amplitude of the clean speech signal of L-1 frame, its In, described voice signal comprises clean speech signal and reverb signal, and described continuous L frame voice signal is frequency-region signal, described L More than 1；

The 1st frame according to described i-th and the continuous L frame voice signal of other M-1 voice-input device input is to L-th frame The amplitude of voice signal, and the continuous L frame voice signal of described i-th and other M-1 voice-input device input 1st frame to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and the coefficient of the clean speech signal amplitude of L-th frame, Clean speech signal to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to L-1 frame The coefficient of the clean speech signal amplitude of reverberation amplitude coefficient and L-th frame is updated；

1st frame of the continuous L frame voice signal according to the input of described i-th voice-input device is to the clean speech of L-1 frame The 1st frame after the amplitude of signal, and the renewal of described i-th voice-input device is to clean speech signal mixed of L-1 frame Ring range coefficient, estimate pure to L-1 frame of the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs The clean voice signal reverberation amplitude to the reverb signal that described L-th frame voice signal produces；

The amplitude of the L-th frame signal of the continuous L frame voice signal according to the input of described i-th voice-input device and described reverberation Amplitude, it is thus achieved that the amplitude of the clean speech signal of described L-th frame signal.

7. the device eliminating reverberation, it is characterised in that this device includes: the first acquisition unit, the first updating block, first Evaluation unit and second obtains unit；

Described first obtains unit, for obtaining i-th and the continuous L frame voice of other M-1 voice-input device input 1st frame of signal is to the energy of the voice signal of L-th frame, and described i-th and other M-1 voice-input device input 1st frame of L frame voice signal is believed to the reverberation energy coefficient of the clean speech signal of L-1 frame and the voice of L-th frame continuously The coefficient of number energy, and pure to L-1 frame of the 1st frame of the continuous L frame voice signal of i-th voice-input device input The energy of voice signal, wherein, described voice signal comprises clean speech signal and reverb signal, and described continuous L frame voice is believed Number being frequency-region signal, described L is more than 1；

Described first updating block, for obtaining, according to described first, described i-th and other M-1 voice that unit sends 1st frame of the continuous L frame voice signal of input equipment input to the energy of the voice signal of L-th frame, and described i-th and its 1st frame of the continuous L frame voice signal of his M-1 voice-input device input is to the reverberation of the clean speech signal of L-1 frame The coefficient of the speech signal energy of energy coefficient and L-th frame, the continuous L frame language to the input of described i-th voice-input device 1st frame of tone signal to the reverberation energy coefficient of the clean speech signal of L-1 frame and L-th frame speech signal energy be Number is updated；

Described first evaluation unit, for obtaining, according to described first, the described i-th voice-input device input that unit sends The 1st frame of continuous L frame voice signal to the energy of the clean speech signal of L-1 frame, and described first updating block sends Described i-th voice-input device renewal after the 1st frame to the reverberation energy coefficient of the clean speech signal of L-1 frame, Estimate the 1st frame clean speech signal to L-1 frame of the continuous L frame voice signal that described i-th voice-input device inputs Reverberation energy to the reverb signal that described L-th frame voice signal produces；

Described second obtains unit, for obtaining, according to described first, the described i-th voice-input device input that unit sends Continuous L frame voice signal L-th frame signal energy and described first evaluation unit send described reverberation energy, it is thus achieved that institute State the energy of the clean speech signal of L-th frame signal.

Device the most according to claim 7, it is characterised in that described first obtain unit specifically for:

The slope q of the Energy Decay Curve according to default impulse signal, by the first formula, it is thus achieved that described i-th and its 1st frame of the continuous L frame voice signal of his M-1 voice-input device input is to the reverberation of the clean speech signal of L-1 frame Energy coefficient g₁、g₂…g_L-1And the coefficient g of the speech signal energy of L-th frame_L, wherein, described first formula is:

Wherein, 1 ＜=l ＜=L.

9. according to the device described in claim 7 or 8, it is characterised in that described L is that reverberation time T is long with the Preset Time of frame The ratio of degree rounds, wherein, and the slope q of the Energy Decay Curve of described default impulse signal according to described reverberation time T, The energy attenuation of the described impulse signal estimated is to time of preset energy.

10. according to the device described in claim 7 or 8, it is characterised in that described first updating block specifically for:

The energy of the continuous L frame signal of i-th voice-input device and jth voice-input device is obtained according to the second formula Difference, wherein, described second formula is:

Wherein, as 1 ＜=l ＜=L-1, described g_j,lE_i,lSet for jth phonetic entry The volume of the signal energy of the reverberation energy coefficient of the clean speech signal of standby l frame and the l frame of i-th voice-input device Long-pending, described g_i,lE_j,lReverberation energy coefficient and the jth language of clean speech signal for the l frame of i-th voice-input device The convolution of the signal energy of the l frame of sound input equipment, as l=L, described g_j,lE_i,lL for jth voice-input device The convolution of the signal energy of the L-th frame of the coefficient of the speech signal energy of frame and i-th voice-input device, described g_i,lE_j,lFor The signal of the L-th frame of the coefficient of the speech signal energy of the L-th frame of i-th voice-input device and jth voice-input device The convolution of energy；

Described i-th voice-input device is obtained with other M-1 voice-input device between any two according to described second formula The difference of energy of continuous L frame signal；

Energy according to described i-th voice-input device He other M-1 voice-input device continuous L frame signal between any two The difference of amount, it is thus achieved that the 3rd formula, described 3rd formula is:

g’_i,l=g_i,l±μ_l* h, wherein, as 1 ＜=l ＜=L-1, described g '_i,lPure for i-th voice-input device l frame The updated value of the reverberation energy coefficient of clean voice signal, described g_i,lFor i-th voice-input device l frame clean speech signal Reverberation energy coefficient, described μ_lThe renewal coefficient of the l frame for pre-seting, described h (e_i, E_l) it is according to i-th phonetic entry The difference of the energy of equipment and other M-1 voice-input device continuous L frame signal between any two, and other M-1 voice The value that 1st frame of the continuous L frame voice signal of input equipment to the energy of the voice signal of L-th frame obtains；

According to described 3rd formula, it is thus achieved that the 1st frame of the described continuous L frame voice signal of described i-th voice-input device is extremely The updated value of the coefficient of the reverberation energy coefficient of the clean speech signal of L-1 frame and the speech signal energy of L-th frame.

11. devices according to claim 10, it is characterised in that described μ_l＞=μ_l-1, wherein, described μ_lFor pre-set The renewal coefficient of l frame, described μ_l-1The renewal coefficient of the l-1 frame for pre-seting, described l-1 frame is the previous of l frame Frame, described 1 ＜=l-1.

12. 1 kinds of devices eliminating reverberation, it is characterised in that this device includes: the 3rd acquisition unit, the second updating block, the Two evaluation units and the 4th acquisition unit:

Described 3rd obtains unit, for obtaining i-th and the continuous L frame voice of other M-1 voice-input device input 1st frame of signal is to the amplitude of the voice signal of L-th frame, and described i-th and other M-1 voice-input device input Continuously the 1st frame of the L frame voice signal reverberation amplitude coefficient to the clean speech signal of L-1 frame and the pure language of L-th frame The coefficient of tone signal amplitude, and the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to L-1 frame The amplitude of clean speech signal, wherein, described voice signal comprises clean speech signal and reverb signal, described continuous L frame language Tone signal is frequency-region signal, and described L is more than 1；

Described second updating block, for obtaining, according to the described 3rd, described i-th and other M-1 voice that unit sends 1st frame of the continuous L frame voice signal of input equipment input to the amplitude of the voice signal of L-th frame, and described i-th and its 1st frame of the continuous L frame voice signal of his M-1 voice-input device input is to the reverberation of the clean speech signal of L-1 frame The coefficient of the clean speech signal amplitude of range coefficient and L-th frame, the continuous L to the input of described i-th voice-input device 1st frame of frame voice signal is to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and the clean speech signal of L-th frame The coefficient of amplitude is updated；

Described second evaluation unit, for obtaining, according to the described 3rd, the described i-th voice-input device input that unit sends The 1st frame of continuous L frame voice signal to the amplitude of the clean speech signal of L-1 frame, and described second updating block sends Described i-th voice-input device renewal after the 1st frame to the reverberation amplitude coefficient of the clean speech signal of L-1 frame, Estimate the 1st frame clean speech signal to L-1 frame of the continuous L frame voice signal that described i-th voice-input device inputs Reverberation amplitude to the reverb signal that described L-th frame voice signal produces；

Described 4th obtains unit, for obtaining, according to the described 3rd, the described i-th voice-input device input that unit sends Continuous L frame voice signal L-th frame signal amplitude and described second evaluation unit send described reverberation amplitude, it is thus achieved that institute State the amplitude of the clean speech signal of L-th frame signal.

13. 1 kinds of equipment eliminating reverberation, it is characterised in that this equipment includes: first transceiver and first processor；

Described first transceiver is used for: receive i-th voice-input device and the company of other M-1 voice-input device input Continuous L frame voice signal；

Described first processor is used for: obtain the continuous L frame voice letter of i-th and other M-1 voice-input device input Number the 1st frame to the energy of the voice signal of L-th frame, and described i-th and the company of other M-1 voice-input device input 1st frame of continuous L frame voice signal is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the voice signal of L-th frame The coefficient of energy, and the 1st frame of the continuous L frame voice signal of i-th voice-input device input is to the pure language of L-1 frame The energy of tone signal, wherein, described voice signal comprises clean speech signal and reverb signal, described continuous L frame voice signal For frequency-region signal, described L is more than 1；

The 1st frame according to described i-th and the continuous L frame voice signal of other M-1 voice-input device input is to L-th frame The energy of voice signal, and the continuous L frame voice signal of described i-th and other M-1 voice-input device input 1st frame is to the reverberation energy coefficient of the clean speech signal of L-1 frame and the coefficient of the speech signal energy of L-th frame, to institute State the 1st frame reverberation to the clean speech signal of L-1 frame of the continuous L frame voice signal that i-th voice-input device inputs The coefficient of the speech signal energy of energy coefficient and L-th frame is updated；

1st frame of the continuous L frame voice signal according to the input of described i-th voice-input device is to the clean speech of L-1 frame The 1st frame after the energy of signal, and the renewal of described i-th voice-input device is to clean speech signal mixed of L-1 frame Ring energy coefficient, estimate pure to L-1 frame of the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs The clean voice signal reverberation energy to the reverb signal that described L-th frame voice signal produces；

The energy of the L-th frame signal of the continuous L frame voice signal according to the input of described i-th voice-input device and described reverberation Energy, it is thus achieved that the energy of the clean speech signal of described L-th frame signal.

14. equipment according to claim 13, it is characterised in that described first processor specifically for:

The slope q of the Energy Decay Curve according to default impulse signal, by the first formula, it is thus achieved that described i-th and its 1st frame of the continuous L frame voice signal of his M-1 voice-input device input is to the reverberation of the clean speech signal of L-1 frame Energy coefficient g₁、g₂…g_L-1And the coefficient g of the speech signal energy of L-th frame_L, wherein, described first formula is:

Wherein, 1 ＜=l ＜=L.

15. according to the equipment described in claim 13 or 14, it is characterised in that described L is the Preset Time of reverberation time T and frame The ratio of length rounds, wherein, and the slope of the Energy Decay Curve of described default impulse signal according to described reverberation time T Q, the time of the energy attenuation of the described impulse signal of estimation to preset energy.

16. according to the equipment described in claim 13 or 14, it is characterised in that described first processor specifically for:

The energy of the continuous L frame signal of i-th voice-input device and jth voice-input device is obtained according to the second formula Difference, wherein, described second formula is:

Wherein, as 1 ＜=l ＜=L-1, described g_j,lE_i,lSet for jth phonetic entry The volume of the signal energy of the reverberation energy coefficient of the clean speech signal of standby l frame and the l frame of i-th voice-input device Long-pending, described g_i,lE_j,lReverberation energy coefficient and the jth language of clean speech signal for the l frame of i-th voice-input device The convolution of the signal energy of the l frame of sound input equipment, as l=L, described g_j,lE_i,lL for jth voice-input device The convolution of the signal energy of the L-th frame of the coefficient of the speech signal energy of frame and i-th voice-input device, described g_i,lE_j,lFor The signal of the L-th frame of the coefficient of the speech signal energy of the L-th frame of i-th voice-input device and jth voice-input device The convolution of energy；

Described i-th voice-input device is obtained with other M-1 voice-input device between any two according to described second formula The difference of energy of continuous L frame signal；

Energy according to described i-th voice-input device He other M-1 voice-input device continuous L frame signal between any two The difference of amount, it is thus achieved that the 3rd formula, described 3rd formula is:

g’_i,l=g_i,l±μ_l* h, wherein, as 1 ＜=l ＜=L-1, described g '_i,lPure for i-th voice-input device l frame The updated value of the reverberation energy coefficient of clean voice signal, described g_i,lFor i-th voice-input device l frame clean speech signal Reverberation energy coefficient, described μ_lThe renewal coefficient of the l frame for pre-seting, described h (e_i, E_l) it is according to i-th phonetic entry The difference of the energy of equipment and other M-1 voice-input device continuous L frame signal between any two, and other M-1 voice The value that 1st frame of the continuous L frame voice signal of input equipment to the energy of the voice signal of L-th frame obtains；

According to described 3rd formula, it is thus achieved that the 1st frame of the described continuous L frame voice signal of described i-th voice-input device is extremely The updated value of the coefficient of the reverberation energy coefficient of the clean speech signal of L-1 frame and the speech signal energy of L-th frame.

17. equipment according to claim 16, it is characterised in that described μ_l＞=μ_l-1, wherein, described μ_lFor pre-set The renewal coefficient of l frame, described μ_l-1The renewal coefficient of the l-1 frame for pre-seting, described l-1 frame is the previous of l frame Frame, described 1 ＜=l-1.

18. 1 kinds of equipment eliminating reverberation, it is characterised in that this equipment includes: second transceiver and the second processor；

Described second transceiver is used for: receive i-th voice-input device and the company of other M-1 voice-input device input Continuous L frame voice signal；

Described second processor is used for: obtain the continuous L frame voice letter of i-th and other M-1 voice-input device input Number the 1st frame to the amplitude of the voice signal of L-th frame, and described i-th and the company of other M-1 voice-input device input 1st frame of continuous L frame voice signal is to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and the clean speech of L-th frame The coefficient of signal amplitude, and pure to L-1 frame of the 1st frame of the continuous L frame voice signal of i-th voice-input device input The amplitude of clean voice signal, wherein, described voice signal comprises clean speech signal and reverb signal, described continuous L frame voice Signal is frequency-region signal, and described L is more than 1；

The 1st frame according to described i-th and the continuous L frame voice signal of other M-1 voice-input device input is to L-th frame The amplitude of voice signal, and the continuous L frame voice signal of described i-th and other M-1 voice-input device input 1st frame to the reverberation amplitude coefficient of the clean speech signal of L-1 frame and the coefficient of the clean speech signal amplitude of L-th frame, Clean speech signal to the 1st frame of the continuous L frame voice signal of described i-th voice-input device input to L-1 frame The coefficient of the clean speech signal amplitude of reverberation amplitude coefficient and L-th frame is updated；

1st frame of the continuous L frame voice signal according to the input of described i-th voice-input device is to the clean speech of L-1 frame The 1st frame after the amplitude of signal, and the renewal of described i-th voice-input device is to clean speech signal mixed of L-1 frame Ring range coefficient, estimate pure to L-1 frame of the 1st frame of the continuous L frame voice signal that described i-th voice-input device inputs The clean voice signal reverberation amplitude to the reverb signal that described L-th frame voice signal produces；

The amplitude of the L-th frame signal of the continuous L frame voice signal according to the input of described i-th voice-input device and described reverberation Amplitude, it is thus achieved that the amplitude of the clean speech signal of described L-th frame signal.