CN106898359A - Acoustic signal processing method, system, audio interactive device and computer equipment - Google Patents

Acoustic signal processing method, system, audio interactive device and computer equipment Download PDF

Info

Publication number
CN106898359A
CN106898359A CN201710182290.XA CN201710182290A CN106898359A CN 106898359 A CN106898359 A CN 106898359A CN 201710182290 A CN201710182290 A CN 201710182290A CN 106898359 A CN106898359 A CN 106898359A
Authority
CN
China
Prior art keywords
signal
audio
noise
residual
audio input
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201710182290.XA
Other languages
Chinese (zh)
Other versions
CN106898359B (en
Inventor
余世经
朱频频
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Zhizhen Intelligent Network Technology Co Ltd
Original Assignee
Shanghai Zhizhen Intelligent Network Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shanghai Zhizhen Intelligent Network Technology Co Ltd filed Critical Shanghai Zhizhen Intelligent Network Technology Co Ltd
Priority to CN201710182290.XA priority Critical patent/CN106898359B/en
Publication of CN106898359A publication Critical patent/CN106898359A/en
Application granted granted Critical
Publication of CN106898359B publication Critical patent/CN106898359B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
    • G10L21/00Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
    • G10L21/02Speech enhancement, e.g. noise reduction or echo cancellation
    • G10L21/0208Noise filtering
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04MTELEPHONIC COMMUNICATION
    • H04M9/00Arrangements for interconnection not involving centralised switching
    • H04M9/08Two-way loud-speaking telephone systems with means for conditioning the signal, e.g. for suppressing echoes for one or both directions of traffic
    • H04M9/082Two-way loud-speaking telephone systems with means for conditioning the signal, e.g. for suppressing echoes for one or both directions of traffic using echo cancellers
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
    • G10L21/00Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
    • G10L21/02Speech enhancement, e.g. noise reduction or echo cancellation
    • G10L21/0208Noise filtering
    • G10L21/0216Noise filtering characterised by the method used for estimating noise
    • G10L21/0232Processing in the frequency domain
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
    • G10L21/00Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
    • G10L21/02Speech enhancement, e.g. noise reduction or echo cancellation
    • G10L21/0208Noise filtering
    • G10L21/0264Noise filtering characterised by the type of parameter measurement, e.g. correlation techniques, zero crossing techniques or predictive techniques
    • GPHYSICS
    • G10MUSICAL INSTRUMENTS; ACOUSTICS
    • G10LSPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
    • G10L21/00Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
    • G10L21/02Speech enhancement, e.g. noise reduction or echo cancellation
    • G10L21/0208Noise filtering
    • G10L2021/02082Noise filtering the noise being echo, reverberation of the speech

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Computational Linguistics (AREA)
  • Quality & Reliability (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Telephone Function (AREA)
  • Circuit For Audible Band Transducer (AREA)

Abstract

The embodiment of the invention discloses acoustic signal processing method, system, audio interactive device and computer equipment.The method includes:Echo cancellor is carried out to the echo signal in acquired audio input signal, the audio residual signals of the audio input signal are obtained;Noise elimination is carried out to the noise signal in the audio residual signals, the audio enhancement signal after noise is eliminated is obtained;Residual noise signal detection is carried out to the audio enhancement signal, if there is residual noise signal in the audio enhancement signal, the residual noise signal is filtered.Using the method, echo signal and ambient noise signal in audio input signal when audio is interacted can be removed, the detection and filtering to residual noise are realized simultaneously, effectively inhibit influence of the ambient noise to audio interaction, the eradicating efficacy of echo and noise in audio interaction is more preferably improved, so as to improve audio interactive device to echo and the elimination performance of noise.

Description

Acoustic signal processing method, system, audio interactive device and computer equipment
Technical field
The present embodiments relate to Audio Signal Processing technical field, more particularly to acoustic signal processing method, system, sound Frequency interactive device and computer equipment.
Background technology
For the intelligent audio interactive device for possessing voice interactive function, it both can receive user by microphone Phonetic order, feedback information can be reported to user by loudspeaker again.But, loudspeaker is reported to the sound of user Can be collected by microphone again as acoustic echo simultaneously, thus influence identification and understanding of the interactive device to user speech, So as to reduce the service behaviour of audio interactive device.Additionally, in actual applications, due to the hardware knot of intelligent audio interactive device Structure is different, and the distance and mounting means of microphone and loudspeaker are different, and the place of system application is also varied, echo path and The time delay situation of echo can also have nothing in common with each other, and these factors both increase the difficulty and complexity of Echo Cancellation Problem.Therefore, such as What effectively eliminates the acoustic echo produced when audio is interacted turns into the problem that technical staff needs to solve.
At present, general that Echo Cancellation Problem is solved using the method for adaptive-filtering, Fig. 1 gives existing based on certainly Adaptive filtering carries out the general principle figure of echo cancellor, specifically, by Fig. 1 it can be found that loudspeaker signal x is by echo path W forms echo signal y, echo signal y and user voice signal v and together gathers to form loudspeaker by the loudspeaker of electronic equipment Signal d;Sef-adapting filter in Fig. 1 can be by its filter coefficient vector of adaptive updatesSo thatContinuous mould Fit and approach echo path w, and according toLoudspeaker signal x treatment to being input into, the analog echo letter that will be exported afterwards NumberSubtracted from loudspeaker signal d, finally give error signal e, user voice signal will be mainly included in the error signal e V, its echo signal is basically eliminated.Have in view of real-time time delay of the voice interactive system to phonetic entry in electronic equipment It is certain to require, therefore, estimation can be segmented when realizing echo cancellor and calculates filter coefficient vector, conventional computational methods are segmentations Block frequency domain adaptive-filtering (Partitioned Block Frequency Domain Adaptive Filter, PBFDAF) Algorithm.
Based on the echo cancellation technology of PBFDAF algorithms, it is proved to be able in the networking telephone and webpage real-time Communication for Power effectively Elimination echo, improve the quality of voice communication, however, performance on the echo cancellor of intelligent audio interactive device is but not to the utmost Such as people's will.First, although PBFDAF algorithms can reduce the echo of loudspeaker generation, for can equally disturb intelligent sound Interactive ambient noise, there is no effect;Secondly as use environment and echo path are changeable, interactive voice scene is each The influence of the factor such as different, adaptive filter algorithm is difficult to for the evaluated error of wave filter to converge to preferable situation, therefore to returning The eradicating efficacy of sound and noise is unable to reach the requirement of intelligent audio interactive device.
The content of the invention
Acoustic signal processing method, system, audio interactive device and computer equipment are the embodiment of the invention provides, more preferably Inhibit influence of the ambient noise to intelligent audio interactive device, improve intelligent audio interactive device noise eliminate property Energy.
In a first aspect, a kind of acoustic signal processing method is the embodiment of the invention provides, including:
Echo cancellor is carried out to the echo signal in acquired audio input signal, the sound of the audio input signal is obtained Frequency residual signals;
Noise elimination is carried out to the noise signal in the audio residual signals, the audio enhancing letter after noise is eliminated is obtained Number;
Residual noise signal detection is carried out to the audio enhancement signal, if there is residual in the audio enhancement signal Noise signal, then filter to the residual noise signal.
Second aspect, the embodiment of the invention provides a kind of audio signal processing, including:
Echo signal cancellation module, for carrying out echo cancellor to the echo signal in acquired audio input signal, obtains Obtain the audio residual signals of the audio input signal;
Noise signal cancellation module, for carrying out noise elimination to the noise signal in the audio residual signals, obtains Audio enhancement signal after noise elimination;
Residual noise detection module, for carrying out residual noise signal detection to the audio enhancement signal;
Residual noise filtering module, for when there is residual noise signal in the audio enhancement signal, to described residual Noise signal is stayed to be filtered.
The third aspect, the embodiment of the present invention additionally provides a kind of audio interactive device, and the audio interactive device includes this The audio signal processing that invention above-described embodiment is provided.
Fourth aspect, the embodiment of the present invention also provides a kind of computer equipment, and the equipment includes:
One or more processors;
Storage device, for storing one or more programs,
One or more of programs are by one or more of computing devices so that one or more of processors Realize the acoustic signal processing method that the above embodiment of the present invention is provided.
5th aspect, the embodiment of the present invention provides a kind of storage medium comprising computer executable instructions again, described Computer executable instructions by computer processor when being performed for performing the audio signal that the above embodiment of the present invention is provided Processing method.
The embodiment of the invention provides acoustic signal processing method, system, audio interactive device and computer equipment, the sound Frequency signal processing method carries out echo cancellor to the echo signal in acquired audio input signal first, obtains the audio input The audio residual signals of signal;Then noise elimination is carried out to the noise signal in the audio residual signals, noise is obtained and is eliminated Audio enhancement signal afterwards;Finally residual noise detection is carried out to audio enhancement signal, and detecting residual noise signal When, the residual noise signal is filtered.Using the method, returning in audio input signal when audio is interacted effectively is eliminated Acoustical signal and ambient noise signal, while realizing detection and filtering to residual noise, compared with the conventional method, effectively suppress Influence of the ambient noise to audio interaction, more preferably improves the eradicating efficacy of echo and noise in audio interaction, So as to improve audio interactive device to echo and the elimination performance of noise.
Brief description of the drawings
Fig. 1 gives the existing general principle figure that echo cancellor is carried out based on adaptive-filtering;
Fig. 2 is a kind of schematic flow sheet of acoustic signal processing method that the embodiment of the present invention one is provided;
Fig. 3 is a kind of schematic flow sheet of acoustic signal processing method that the embodiment of the present invention two is provided;
Fig. 4 a are a kind of schematic flow sheet of acoustic signal processing method that the embodiment of the present invention three is provided;
Fig. 4 b are that a kind of acoustic signal processing method provided based on the embodiment of the present invention three carries out Audio Signal Processing General principle figure;
Fig. 5 a are the schematic flow sheet of the preferred embodiment of the acoustic signal processing method that the embodiment of the present invention four is provided;
Fig. 5 b are the oscillogram of the preferred embodiment sound intermediate frequency input signal that the embodiment of the present invention four is provided;
Fig. 5 c are the oscillogram of the preferred embodiment sound intermediate frequency reference signal that the embodiment of the present invention four is provided;
The treatment effect processed audio input signal in the preferred embodiment that Fig. 5 d are provided for the embodiment of the present invention four Fruit is schemed;
Fig. 5 e are the treatment effect processed audio input signal based on a kind of existing acoustic signal processing method Figure;
Fig. 5 f are the treatment effect processed audio input signal based on existing another acoustic signal processing method Fruit is schemed;
Fig. 6 is a kind of structured flowchart of audio signal processing that the embodiment of the present invention five is provided;
Fig. 7 is a kind of structured flowchart of computer equipment that the embodiment of the present invention six is provided.
Specific embodiment
The present invention is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched The specific embodiment stated is used only for explaining the present invention, rather than limitation of the invention.It also should be noted that, in order to just Part rather than entire infrastructure related to the present invention is illustrate only in description, accompanying drawing.
Embodiment one
Fig. 2 is a kind of schematic flow sheet of acoustic signal processing method that the embodiment of the present invention one is provided, and the method is applicable In the audio interactive device with audio interactive function carry out audio interact when to pick up audio input signal carry out echo and The situation of noise processed, the method can be performed by audio signal processing, and wherein the system can be by software and/or hardware reality It is existing, and be typically integrated on the audio interactive device with audio interactive function.
It should be noted that the voice that the audio interaction specifically can be regarded as between user and audio interactive device is handed over Mutually, the audio interactive device can be that intelligent sound assistant, smart home and Voice Navigation instrument etc. carry interactive voice work( The electronic product of energy.
As shown in Fig. 2 the acoustic signal processing method that the embodiment of the present invention one is provided, including following operation:
S101, echo cancellor is carried out to the echo signal in acquired audio input signal, obtain the audio input letter Number audio residual signals.
In the present embodiment, audio input signal can be obtained in units of frame from input-buffer area set in advance, Wherein, the input-buffer area is set in advance in audio interactive device, and input-buffer area sound intermediate frequency input signal can be in sound Picked up by the audio input device on audio interactive device during frequency interaction, the audio input device can specifically refer to that audio is interacted Microphone and microphone in device etc. are for picking up the audio pickup device of audio signal data.
Specifically, when carrying out audio interaction, echo signal has been frequently included in acquired audio input signal and has been made an uproar Acoustical signal etc., the audio reference signal that the echo signal specifically can be regarded as audio output apparatus output is passed through in external environment The signal picked up by audio input device after one or many reflection is crossed, wherein, the audio output apparatus can specifically refer to audio Receiver and loudspeaker in interactive device etc. are for the audio player part of playing audio-fequency data, and the audio reference signal is specific The signal of audio output apparatus output can be referred to;The noise signal specifically can be regarded as that audio input device picked up except user Other signals outside voice signal and echo signal.
This step carries out echo cancellor by automatic echo cancellor strategy to the echo signal in audio input signal first Treatment, is derived from corresponding to the audio residual signals of the audio input signal, and the audio residual signals are specifically appreciated that It is the signal formed after being eliminated to the echo signal in audio input signal.
In the present embodiment, adoptable automatic echo cancellor strategy includes:Based on the adaptive of canonical least mean-square error Filtering algorithm and PBFDAF algorithms etc. are answered, the operation principle that above-mentioned algorithm carries out echo cancellor is:Above-mentioned algorithm can lead to The echo path for crossing sef-adapting filter environmental test forms the echo close with echo signal in audio input signal and estimates Meter signal, formed echo may finally be subtracted from audio input signal and estimates signal, equivalent to balancing out audio input Echo signal in signal, the elimination for being achieved in echo signal obtains audio residual signals.
S102, noise elimination is carried out to the noise signal in the audio residual signals, obtain the audio after noise is eliminated Enhancing signal.
This step can carry out noise elimination after echo cancellor to the noise signal in audio residual signals, specifically (wherein, S represents audio residual signals to the signal to noise ratio S/N of ground, predetermined noise signal and audio residual signals, and N represents audio Noise signal in residual signals), and noise signal corresponding decay system in audio residual signals is determined according to signal to noise ratio S/N Number, the attenuation coefficient has been specifically indicated that damping ratios of the noise signal in audio residual signals, thus can have been declined by described Subtracting coefficient carries out the decay of noise signal with the product of the audio residual signals, to realize the elimination of noise signal.Can manage Solution, noise signal eliminate after audio residual signals relative to realizing signal enhancing before elimination, therefore can be by noise Audio residual signals after elimination are referred to as audio enhancement signal.
S103, residual noise signal detection is carried out to the audio enhancement signal, if deposited in the audio enhancement signal In residual noise signal, then the residual noise signal is filtered.
This step can to obtain audio enhancement signal further carry out residual noise detection, it is possible to it is determined that in the presence of Filtration treatment is carried out to residual noise signal during residual noise signal.Specifically, by audio input signal and sound can be obtained The ratio between energy value of frequency enhancing signal is defined as the parameter values for detection of residual noise signal, while the detection parameter that will can be determined Value is compared with the detection threshold value of setting, and the audio enhancing letter is determined when the parameter values for detection is more than detection threshold value There is residual noise signal in number, wherein, the detection threshold value can specifically set according to historical experience value.
In the present embodiment, if it is determined that there is residual noise signal in audio enhancement signal, then can determine that residual is made an uproar Acoustical signal corresponding attenuation coefficient in audio enhancement signal.Specifically, the detection threshold value can be determined with the inspection first The ratio of parameter value is surveyed, then the ratio and the size for setting drop threshold, the final maximum chosen in the two is made Attenuation coefficient for residual noise signal relative to audio enhancement signal.Wherein, the attenuation coefficient also show residual noise letter Damping ratios number in audio enhancement signal, thus can be entered by the attenuation coefficient with the product of the audio enhancement signal The decay of row residual noise signal, to realize the filtering of residual noise signal.
A kind of acoustic signal processing method that the embodiment of the present invention one is provided, first in acquired audio input signal Echo signal carries out echo cancellor, obtains the audio residual signals of the audio input signal;Then in the audio residual signals Noise signal carry out noise elimination, obtain the audio enhancement signal after noise is eliminated;Finally audio enhancement signal is carried out residual Noise measuring is stayed, and when residual noise signal is detected, the residual noise signal is filtered.Using the method, effectively Echo signal and ambient noise signal in audio input signal when audio is interacted are eliminated, while realizing to residual noise Detection and filtering, compared with the conventional method, the method effectively prevent influence of the ambient noise to echo cancellation convergence result, also Influence of the ambient noise to audio interaction effectively is inhibited, disappearing for echo and noise in audio interaction is more preferably improved Except effect, and then improve the noise removing performance of audio interactive device.
Embodiment two
Fig. 3 is a kind of schematic flow sheet of acoustic signal processing method that the embodiment of the present invention two is provided, and the present invention is implemented Example is optimized based on above-described embodiment, in the present embodiment, the noise signal in the audio residual signals will be entered Row noise is eliminated, and obtains the audio enhancement signal after noise is eliminated, and is further optimized for:According to the work(of the audio residual signals The noise power spectrum of noise signal, determines the audio residual signals in rate spectrum and the predetermined audio residual signals The first posteriori SNR and prior weight;According to first posteriori SNR and the prior weight, it is determined that described First attenuation coefficient of noise signal in audio residual signals;The audio residual signals are eliminated according to first attenuation coefficient In noise signal, obtain the audio enhancement signal of the audio residual signals.
The present embodiment further will carry out residual noise signal detection to the audio enhancement signal, if the audio There is residual noise signal in enhancing signal, then the residual noise signal is filtered and be specifically optimized for:According to the sound Frequency input signal, the audio enhancement signal and default detection parameter equation, determine that the residual of the audio enhancement signal is made an uproar Sound detection parameter;If the residual noise detection parameter is less than the detection parameter threshold of setting, it is determined that the audio enhancing There is residual noise signal in signal;Parameter and the detection parameter threshold are detected according to the residual noise, obtains described Second attenuation coefficient of the residual noise signal in the audio enhancement signal;According to second attenuation coefficient, filtering is described Residual noise signal in audio enhancement signal.
Additionally, the present embodiment will also carry out echo cancellor to the echo signal in acquired audio input signal, institute is obtained The audio residual signals of audio input signal are stated, is specifically optimized for:Obtain audio input device pickup audio input signal with And the audio reference signal of audio output apparatus output;The audioref according to the current adaptive-filtering coefficient processing for determining Signal, the echo for obtaining the audio reference signal estimates signal;The audio input signal is estimated into signal with the echo Difference be defined as the audio residual signals of the audio input signal.
As shown in figure 3, a kind of acoustic signal processing method that the embodiment of the present invention two is provided, specifically includes following operation:
In the present embodiment, S201~S203 gives the concrete operation step of echo cancellor, and the present embodiment is preferably adopted Echo cancellor is carried out with PBFDAF algorithms.
The audioref of S201, the audio input signal for obtaining audio input device pickup and audio output apparatus output Signal.
Specifically, this step obtains the audio input signal of current pending echo cancellor, disappears while obtaining and carrying out echo Except required audio reference signal, wherein, the present embodiment can obtain audio from the reference buffer storage area of setting in units of frame Reference signal, acquired audio reference signal can be used for audio input signal as the input signal of sef-adapting filter Echo cancellor.
It should be noted that the audio that the echo signal in audio input signal refers specifically to be exported by audio output apparatus is joined Signal of the signal by formation after the echo path of external environment is examined, meanwhile, the self adaptation filter according to described in above-described embodiment Ripple device carries out the operation principle of echo cancellor, and the present embodiment can obtain audio reference signal, and audio reference signal is passed through Sef-adapting filter forms analog echo path, and signal is estimated to generate the echo close with actual echo signal.Additionally, described Reference buffer storage area is set in advance in audio interactive device, is believed for caching the audioref for treating to export by audio output apparatus Number.
S202, the audio reference signal according to the current adaptive-filtering coefficient processing for determining, obtain the audio ginseng The echo for examining signal estimates signal.
This step uses PBFDAF algorithm process audio reference signals, is derived from the corresponding echo of the audio reference signal Estimate signal, specifically, its echo estimates that the determination process of signal is expressed as:The audio reference signal that will be obtained is used as PBFDAF The input value of algorithm, the echo signal path of current environment is simulated according to the current adaptive-filtering coefficient for determining, and then in frequency Treatment is filtered on rate domain to audio reference signal, the echo close with actual echo signal is obtained and is estimated signal.
In the present embodiment, above-mentioned adaptive-filtering coefficient specifically can be regarded as working as front ring for simulation in adaptive-filtering The filter factor of border echo path.Usually, initial adaptive-filtering coefficient can be initially set 0 vector, can return afterwards Sound determines updating factor while elimination, and the renewal of adaptive-filtering coefficient is carried out according to the updating factor for determining, wherein, institute The specific value for stating updating factor is not fixed, and the audio residual signals obtained when specifically to echo cancellor are related.Can be with Understand, the specific value of adaptive-filtering coefficient of the current determination can be returned to former frame audio input signal Updated during sound Processing for removing and obtained.
It should be noted that initial adaptive-filtering coefficient can also be initially set other any one non-vanishing vectors, It does not influence protection scope of the present invention.
S203, the audio input signal and the echo are estimated that the difference of signal is defined as the audio input signal Audio residual signals.
This step realizes and the operation that echo estimates signal is eliminated in audio input signal, is derived to audio input Signal eliminates the audio residual signals after echo signal.In the present embodiment, S204~S206 gives noise signal elimination Concrete operation step.
S204, the power spectrum according to the audio residual signals and noise in the predetermined audio residual signals The noise power spectrum of signal, determines first posteriori SNR and prior weight of the audio residual signals.
This step determines the first posteriori SNR and prior weight eliminated for noise signal first.Specifically, originally Step can obtain the power spectrum of audio residual signals, while noise signal in predetermined audio residual signals can be obtained Noise power spectrum, afterwards according to determine audio residual signals power spectrum and noise power spectrum obtain needed for first Posteriori SNR and prior weight.
It should be noted that currently employed noise power spectrum specifically can carry out signal to former frame audio input signal Updated during treatment and obtained, and the initial value of noise power spectrum can be based on the power spectrum of acquired preceding 3 to 5 frame audio residual signals Average determines.Specifically, the initial value of noise power spectrum can be obtained by following formula: Wherein, j represents the frequency number of frequency domain, is any integer value in 0~M-1, and M represents frequency length of field, NE0J () represents The initial value of the corresponding noise power spectrums of frequency number j, i represents the frame number of audio input signal, 0<i<M, m represent calculating noise The frame number of the audio residual signals needed for the initial value of power spectrum, 3≤m≤5, | E (i, j) |2Represent the i-th frame audio input signal Audio residual signals power spectrum.
Further, the present embodiment is obtained according to following first posteriori SNR formula and prior weight formula respectively First posteriori SNR and prior weight of the audio residual signals are obtained, the first posteriori SNR formula is expressed as:The prior weight formula is expressed as:
Wherein, j represents the frequency number of frequency domain, is any integer value in 0~M-1, and M represents frequency length of field, n tables Show the frame number for being presently in managing audio input signal, α is the either constant between 0~1, SNRpost(n, j) represents n-th frame audio First posteriori SNR of the audio residual signals of input signal, SNRprio(n, j) represents the audio of n-th frame audio input signal The prior weight of residual signals, | E (n, j) |2The power spectrum of the audio residual signals of n-th frame audio input signal is represented, | NE (n,j)|2The noise power spectrum of noise signal in the audio residual signals of n-th frame audio input signal is represented, | E1 (n-1, j) |2 Represent the enhancing power spectrum signal of the audio residual signals of the (n-1)th frame audio input signal.
In the present embodiment, | E1 (n-1, j) |2Can be according to the audio residual signals pair of the (n-1)th frame audio input signal The audio enhancement signal answered determines, it is to be understood that the (n-1)th frame audio input signal is equivalent to n-th frame audio input signal Former frame, therefore, the corresponding audio enhancement signal of audio residual signals of the (n-1)th frame audio input signal is equally based on this hair The acoustic signal processing method that bright embodiment is provided determines, and its treatment operation is needed to the treatment of n-th frame audio input signal Carry out before.
S205, according to first posteriori SNR and prior weight, determine noise letter in the audio residual signals Number the first attenuation coefficient.
This step can be according to the first posteriori SNR and prior weight for determining, it is determined that for noise signal elimination First attenuation coefficient, further, the present embodiment is obtained in the audio residual signals according to following first attenuation coefficient formula First attenuation coefficient of noise signal, the first attenuation coefficient formula is expressed as:
Wherein, ζ 1 (n, j) represents the first attenuation coefficient of the audio residual signals of n-th frame audio input signal.
In the present embodiment, in above-mentioned formula exp () is specially exponent arithmetic, and exp int () are specially exponent product Partite transport is calculated.
S206, the noise signal in first attenuation coefficient elimination audio residual signals, obtain the sound The audio enhancement signal of frequency residual signals.
The present embodiment obtains corresponding first attenuation coefficient of audio residual signals by above-mentioned steps, first is declined according to it Subtracting coefficient can carry out noise attentuation to the noise signal in audio residual signals, equivalent to removal audio residual signals in make an uproar Acoustical signal, is derived from audio enhancement signal.
It is possible to further determine that the audio of n-th frame audio input signal is residual according to following enhancing power spectrum signal formula The enhancing power spectrum signal of difference signal, the enhancing power spectrum signal formula is expressed as:
|E1(n,j)|2=ζ 1 (n, j) | E (n, j) |2,
Wherein, | E1 (n, j) |2Represent the enhancing power spectrum signal of the audio residual signals of n-th frame audio input signal.
It is understood that the audio residual signals in frequency domain can be determined based on above-mentioned enhancing power spectrum signal formula Enhancing power spectrum signal, this step can according to the enhancing power spectrum signal determine audio residual signals in frequency domain correspondence Audio enhancement signal, afterwards can pair audio enhancement signal for determining carry out inverse Fourier transform, it is final to obtain in time domain Audio enhancement signal.
In the present embodiment, S207~S210 gives the specific operation process of detection and the filtering of residual noise signal.
S207, according to the audio input signal, the audio enhancement signal and default detection parameter equation, it is determined that The residual noise detection parameter of the audio enhancement signal.
Detection parameter equation of this step first according to audio input signal, the audio enhancement signal for determining and setting determines For the detection parameter of residual noise detection.
Further, the detection parameter equation is expressed as:
Wherein, n represents the frame number for being presently in managing audio input signal, ∑ | d (n, t) |2 Represent the energy value of n-th frame audio input signal;∑|e1(n,t)|2Represent the audio enhancement signal of n-th frame audio input signal Energy value;RD (n) represents the residual noise detection parameter of the audio enhancement signal of n-th frame audio input signal.
If S208, the residual noise detection parameter are less than the detection parameter threshold of setting, it is determined that the audio increases There is residual noise signal in strong signal.
This step can compare to determine audio enhancement signal by residual noise parameter with the size of detection parameter threshold In whether there is residual noise signal, specifically, when residual noise parameter less than detection parameter threshold when, it is determined that being made an uproar in the presence of residual Acoustical signal.It should be noted that the detection parameter threshold occurrence can be manually set according to historical experience value, or according to Specifically used scene determines corresponding value.
S209, parameter and the detection parameter threshold are detected according to the residual noise, obtain residual noise letter The second attenuation coefficient number in the audio enhancement signal.
, it is necessary to further filter the residual noise signal when it is determined that there is residual noise signal, this step determines first The second attenuation coefficient needed for filtering the residual noise signal.
Specifically, the present embodiment can determine the residual noise signal described according to following second attenuation coefficient formula The second attenuation coefficient in audio enhancement signal, the second attenuation coefficient formula is expressed as:
ζ 2 (n)=max (RDth/RD(n),ζmin), wherein, RDthIt is the detection parameter threshold of setting;ζminFor setting most Small attenuation coefficient;Second decay of residual noise signal in the audio enhancement signal of ζ 2 (n) expression n-th frame audio input signals Coefficient.
In the present embodiment, RDthValue be typically set at constant more than 1, ζminValue be typically set at more than 0 And the constant less than 1, i.e. ζmin∈ (0,1), its specific value is preferable to be set according to historical experience value.
S210, according to second attenuation coefficient, filter the residual noise signal in the audio enhancement signal.
In the present embodiment, can be to the residual noise signal in audio enhancement signal again according to its second attenuation coefficient Noise attentuation is carried out, filtration treatment is carried out again equivalent to the noise signal in audio enhancement signal, it is to be understood that its Audio signal after filtration treatment is target audio signal to be output.
Further, after being filtered to the residual noise signal, also include:
Determine formula according to following signals, determine target audio signal to be output;
The signal determines that formula is expressed as:S (n)=ζ 2 (n) * e1 (n), wherein, e1 (n) represents n-th frame audio input The audio enhancement signal of signal, s (n) represents the target audio signal of n-th frame audio input signal to be output.
If it is understood that determining detection parameter of the residual noise detection parameter not less than setting according to above-mentioned S208 Threshold value, then can determine that in audio enhancement signal do not exist residual noise signal, and now, the present embodiment is directly by audio enhancement signal Exported as target audio signal.
It should be noted that the target audio signal of the present embodiment also passes through audio output apparatus output, the target It is set in advance that audio signal and the difference of above-mentioned audio reference signal are that the target audio signal for being obtained is cached in first In reference buffer storage area, reference buffer storage area can regard a queue data structure as, the characteristics of with first in first out, i.e. only target When audio signal reaches output and requires in reference buffer storage area, just exported by audio output apparatus, and the sound in the present embodiment Frequency reference signal can regard the current signal treated from audio output apparatus output in reference buffer storage area as.Generally, it is to be output Audio reference signal and treatment obtain target audio signal between there may be multiframe treating by Audio Signal Processing The signal of output.
A kind of acoustic signal processing method that the embodiment of the present invention two is provided, specifically employs the filter of fragmented blocks frequency domain adaptive Ripple carries out echo cancellor to audio input signal, while embodying the detection for removing division operation and residual noise of ambient noise And filter operation.Using the method, echo signal and environment in audio input signal when audio is interacted can be effectively eliminated Noise signal, while realizing detection and filtering to residual noise.The method also effectively prevent ambient noise and echo disappeared Except when evaluated error restrain result influence, more preferably improve the eradicating efficacy of echo and noise in audio interaction so that Audio interactive device is improve to echo and the elimination performance of noise.
On the basis of above-mentioned optimization, the embodiment of the present invention further the power spectrum according to the audio residual signals with And in the predetermined audio residual signals noise signal noise power spectrum, determine the first of the audio residual signals After posteriori SNR and prior weight, optimization increased:According to first posteriori SNR and the prior weight, Determine that the noise of the audio residual signals estimates regulation parameter;If the noise estimates regulation parameter less than setting regulation ginseng Number threshold value, then update the noise power spectrum and preserve.
In the present embodiment, except based on aforesaid operations audio input signal is carried out echo cancellor, noise eliminate and it is residual Stay outside noise filtering, noise work(required when stage renewal determines treatment next frame audio input signal can also be eliminated in noise Rate is composed, you can operated with the renewal that noise power spectrum is synchronously carried out in the noise elimination stage.
In the present embodiment, the value of the setting regulation parameter threshold value is typically set at the constant more than 0 and less than 1, Its specific value is preferable to be set according to historical experience value.Specifically, it is presently in reason audio input signal correspondence in acquisition Audio residual signals and after determining first posteriori SNR and prior weight of the audio residual signals, can according to its One posteriori SNR and prior weight determine that the noise needed for updating noise power spectrum estimates regulation parameter.
Further, the present embodiment estimates that regulation parameter formula obtains making an uproar for the audio residual signals according to following noises Sound estimates regulation parameter, and the noise estimates that regulation parameter formula is expressed as:
Wherein, β represents that the noise of n-th frame audio residual signals estimates regulation parameter.
In the present embodiment, when β is less than above-mentioned setting regulation parameter threshold value, the update condition of noise power spectrum is met, It is possible to further update the noise power spectrum according to following noise power spectrums more new formula, the noise power spectrum updates Formula is expressed as:|NE(n,j)|2=ε | NE (n-1, j) |2+(1-ε)|E(n,j)|2, wherein, | NE (n-1, j) |2Represent (n-1)th The noise power spectrum of noise signal in the audio residual signals of frame audio input signal.
In the present embodiment, the value of ε is typically set at the constant more than 0 and less than 1, i.e. ε ∈ (0,1), and it specifically takes Value is preferable to be set according to historical experience value, | NE (n-1, j) |2Can be according to the (n-1)th frame audio residual signals and n-th -2 The noise power spectrum of noise signal determines in frame audio residual signals, and its determination result can equally be carried based on the embodiment of the present invention The acoustic signal processing method of confession is realized, and its treatment operation was typically carried out before to the treatment of n-th frame audio input signal.
The renewal of noise power spectrum described in the present embodiment can be used for follow-up audio residual signals and determine more accurate the One posteriori SNR and prior weight, so as to ensure the follow-up more accurate audio enhancement signal of acquisition and export more high-quality Target audio signal.
Embodiment three
Fig. 4 a are a kind of schematic flow sheet of acoustic signal processing method that the embodiment of the present invention three is provided.The present invention is implemented Example is optimized based on above-described embodiment, in the present embodiment, is estimated with the echo by the audio input signal The difference of signal is defined as after the audio residual signals of the audio input signal, and also optimization is included:According to the audio Reference signal and the audio residual signals for determining, determine the updating factor of frequency domain filter coefficient;According to the updating factor Update the adaptive-filtering coefficient and preserve.
Further, the echo signal in acquired audio input signal carries out echo cancellor, obtains the audio After the audio residual signals of input signal, also include:Whether detect in the audio input signal comprising user voice signal, And whether detect in the audio reference signal comprising echo signal;When in the audio input signal include the user speech During signal, stop updating the adaptive-filtering coefficient;Or, when in the audio input signal do not include the user speech When not including echo signal in signal and the audio reference signal, stop updating the adaptive-filtering coefficient;Or, work as institute When stating in audio input signal not comprising in the user voice signal and the audio reference signal comprising echo signal, update The adaptive-filtering coefficient is simultaneously preserved.
The present embodiment following step is specifically described the qualifications and adaptive-filtering of adaptive-filtering coefficient update The renewal operation of coefficient.It is understood that after echo cancellor acquisition audio residual signals are carried out to audio input signal, may be used also Adaptive-filtering coefficient is updated according to audio residual signals are obtained, the adaptive-filtering coefficient after renewal can be used for next frame sound The echo cancellation process of frequency input signal, additionally, the present embodiment can also according to propose update restrictive condition carry out it is adaptive Filter factor is answered before updating to determine whether that the renewal of adaptive-filtering coefficient can be carried out.
As shown in fig. 4 a, a kind of acoustic signal processing method that the embodiment of the present invention three is provided, specifically includes following operation:
The audioref of S301, the audio input signal for obtaining audio input device pickup and audio output apparatus output Signal.
Exemplary, when the acoustic signal processing method based on the present embodiment carries out Audio Signal Processing, can be from setting Input-buffer area in obtain a currently pending frame audio input signal, while one can be obtained from setting reference buffer storage area Frame is used for the audio reference signal of echo cancellor.
S302, the audio reference signal according to the current adaptive-filtering coefficient processing for determining, obtain the audio ginseng The echo for examining signal estimates signal.
Exemplary, the adaptive-filtering coefficient in this step can be carried out at signal to former frame audio input signal Determine during reason, the present embodiment can according to the fixed adaptive-filtering coefficient according to PBFDAF algorithm simulation echo paths, from And obtain the corresponding echo of audio reference signal and estimate signal, finally realize echo cancellor.
In the present embodiment, the process for echo cancellor being carried out to acquired audio input signal based on PBFDAF algorithms can table State for:Assuming that the current adaptive-filtering coefficient for determining is w(n), length is P ranks, wherein, n represents current to join to n-th frame audio Examine signal to be processed, P is a constant, the specific value of P determines that echo delay time is more long by the echo delay time of echo path, P's Value is higher, and exemplarily, when the echo delay time of echo path is 10ms, the value of its P can be 9, i.e. adaptive-filtering The length of coefficient is 9 ranks;When the echo delay time of echo path is 20ms, the value of its P can be 15, i.e. adaptive-filtering system Several length is 15 ranks, additionally, being the timeliness of guarantee echo cancellor, the highest value of P is often not more than 100.Because of the algorithm Core is fragmented blocks, so first can be in time domain by w(n)It is divided into K sections, kth section adaptive-filtering coefficient wk (n)Length It is N=P/K, and wk (n)Specifically it is represented by:
Wherein, k=0,1,2 ..., K-1.
It is ensured that the every section of adaptive-filtering coefficient w for being obtainedk (n)Length be N, if setting frequency domain Length is M, then can be to every section of adaptive-filtering coefficient wk (n)Zero padding is carried out, and can after zero-filling pass through Fast Fourier Transform (FFT) K band frequencies domain filter coefficient, every band frequency domain filter coefficient W are obtained in frequency domaink (n)It is represented by:
Wherein, 0M-NRepresent M-N rank null vectors.
Subsequently can be according to each band frequency domain filter coefficient W for determiningk (n)Echo cancellor is carried out respectively, specifically, first Audio reference signal x (n) for obtaining is divided into k sections, and carries out Fourier transformation respectively, obtain the audioref in k band frequencies domain Signal, the audio reference signal in frequency domain is represented by:
Wherein, L represents acquired unit frame audio reference signal Data length.
In the present embodiment, the duration of unit frame can specifically determine according in concrete scene, but when audio interactive device tool When having fixed sample rate, the duration of unit frame is proportional with data length, typically can be true by the duration span of unit frame It is set to 1ms~100ms.It should be noted that set unit frame duration does not interfere with the final process of audio input signal As a result, and the present embodiment be ensure per frame data length, the duration of unit frame is preferably set as 10ms~30ms.Example Property, if the sample rate of currently used audio interactive device is 16KHz, the data length of the present embodiment audio reference signal Scope is 160/3~160.
It is then possible to the audio reference signal by every band frequency domain filter coefficient respectively to corresponding frequencies domain is filtered Ripple is processed and carries out result and adds up, and can take rear L sampled data for effective echo estimates letter in the result after adding up afterwards Number, exemplarily, its echo estimates signalIt is represented by:
Wherein, 0M-LRepresent M-L rank null matrix, ILRepresent L ranks Unit matrix.
S303, the audio input signal and the echo are estimated that the difference of signal is defined as the audio input signal Audio residual signals, respectively perform S304 and S306.
Above-mentioned example is connect, audio residual signals e (n) can be expressed as:In addition this step is obtained The audio residual signals after echo cancellor are obtained, subsequently can realize that the audio residual signals noise is believed by performing S304 and S305 Number eliminate and residual echo signal detection and filtering, while can also according to S306 and its subsequent step determine whether update Adaptive-filtering coefficient, and the self adaptation filter for next frame audio input signal echo cancellor is updated when update condition is met Wave system number.
S304, noise elimination is carried out to the noise signal in the audio residual signals, obtain the audio after noise is eliminated Enhancing signal.
S305, residual noise signal detection is carried out to the audio enhancement signal, if deposited in the audio enhancement signal In residual noise signal, then the residual noise signal is filtered.
The method that the present embodiment can be illustrated using above-described embodiment realizes S304 and S305, and I will not elaborate.
Whether user voice signal is included in S306, the detection audio input signal, if so, then performing S308;If it is not, Then perform S307.
In the present embodiment, signal transacting is carried out to audio input signal typically in units of frame, if currently processed User voice signal is contained in audio input signal, then stops updating adaptive-filtering coefficient, to avoid follow-up echo cancellor When the echo path of simulation deviate real echo path.Specifically, this step can be it is determined that comprising user voice signal Shi Zhihang S308;S307 can also be performed when it is determined that not including user voice signal.
Further, whether user voice signal is included in the detection audio input signal, including:
The energy value of energy value and the audio reference signal according to the audio input signal, determines that the audio is defeated Enter the instantaneous energy ratio of signal and the audio reference signal;If the instantaneous energy ratio is less than the first given threshold, really User voice signal is not included in the fixed audio input signal;Otherwise, it determines comprising the user voice signal.
In this example, it is assumed that currently processed audio input signal is d (n), audio reference signal is x (n), then sound The energy value of frequency input signal can be expressed as | d (n, t) |2, the energy value of audio reference signal can be expressed as | x (n, t) |2, The instantaneous energy of the two can be expressed as than η (n):
In the present embodiment, first given threshold can be either constant, and its value can be by historical experience value Or sample training determines, and different restrictions can be carried out according to the difference of application scenarios, it is exemplary, carrying out audio interaction During, it is 0.5 that can set first given threshold according to its historical experience value.
Whether echo signal is included in S307, the detection audio reference signal, if so, then performing S309;If it is not, then holding Row S308.
The present embodiment it is determined that not comprising user voice signal when, can further detect whether comprising echo signal, if Not comprising user voice signal and not comprising echo signal in currently processed audio input signal, then it is believed that current audio It is no user voice signal input also echoless signal input in input signal, therefore to avoid the echo path of simulation from deviateing true Real echo path, needs also exist for stopping updating adaptive-filtering coefficient.
Specifically, this step can perform S308 when it is determined that not including echo signal;When can also include echo signal Perform S309.
Further, whether echo signal is included in the detection audio reference signal, including:
Power spectrum and the corresponding noise power spectrum of the audio residual signals according to the audio input signal, determine institute State the second posteriori SNR of audio input signal;Performance number and the audio residual signals according to the audio input signal Performance number, determine the cross-correlation coefficient of the audio input signal and the audio residual signals;If second posteriority Signal to noise ratio is less than the second given threshold, it is determined that do not include echo signal in the audio reference signal;Or, if described Second posteriori SNR is not less than the second given threshold and the cross-correlation coefficient is not less than the 3rd given threshold, it is determined that described Echo path comprising echo signal and the echo signal in audio reference signal there occurs change;Or, if described Two posteriori SNRs are not less than second given threshold and the cross-correlation coefficient is less than the 3rd given threshold, it is determined that Echo path comprising echo signal and the echo signal in the audio reference signal does not change.
In this example, it is assumed that currently processed audio input signal is d (n), corresponding audio residual signals are e (n), the then power spectrum of audio input signal is represented by ∑ in frequency domain | D (n, j) |2, noise signal in audio residual signals Noise power spectrum | the NE (n, j) | that is expressed as ∑2, the second posteriori SNR σ (n) of audio input signal is represented by:
Equally, audio input signal is represented by P for the performance number of d (n)dN (), audio residual signals are the power of e (n) Value is represented by PeN (), cross-correlation coefficient ξ (n) of the two is represented by:
Wherein, Pd,e(0)=e (0) dT(0), Pd,e(n)=λ Pd,e(n-1)+(1-λ)e(n)dT(n),
Pd(0)=d (0) dT(0), Pd(n)=λ Pd(n-1)+(1-λ)d(n)dT(n),
Pe(0)=e (0) eT(0), Pe(n)=λ Pe(n-1)+(1-λ)e(n)eT(n), λ ∈ (0,1).
In the present embodiment, second given threshold and the 3rd given threshold can be either constants, and its value is equal Can be determined by historical experience value or sample training, and different restrictions can be carried out according to the difference of application scenarios, additionally, λ can be the either constant between 0 to 1, can set its specific value according to historical experience, exemplarily, can be according to history It is 1.2 that empirical value sets second given threshold respectively, sets the 3rd given threshold as 1.25, and sets the λ It is 0.75.
S308, stopping update the adaptive-filtering coefficient.
In the present embodiment, when in detecting audio input signal comprising user voice signal, this step can be carried out Operation;Or, not comprising user voice signal and during not comprising echo signal in currently detected audio input signal, equally The operation of this step can be carried out.
Specifically, the renewal that updates adaptive filter coefficient can it is determined that after audio residual signals, suspended, can be still Carry out carrying out echo cancellor to follow-up audio input signal with the current adaptive-filtering coefficient for determining, and only in subsequent sound Self adaptation just can be then updated after being determined for compliance with the update condition of adaptive-filtering coefficient in the treatment operation of frequency input signal Filter factor.
It should be noted that work as to stop updating adaptive-filtering coefficient, and the follow-up update condition that meets again is proceeded by certainly During adaptive filtering coefficient, then currently used adaptive-filtering coefficient can be updated.
S309, the audio residual signals according to the audio reference signal and determination, determine frequency domain filter coefficient Updating factor.
In the present embodiment, when the update condition of adaptive-filtering coefficient is met, can be entered based on this step and S310 The renewal of row adaptive-filtering coefficient.Specifically, believe in the audio residual error that audio input signal is determined according to PBFDAF algorithms After number, so that it may the updating factor of the frequency domain filter coefficient of sef-adapting filter is obtained according to this step.
Connect the example of above-mentioned S303, it is assumed that the current audio reference signal for obtaining is x (n), its be divided into k sections after in frequency Adjoint matrix on domain is expressed as X* k(n), it is assumed that the audio residual signals determined after echo cancellor are e (n), then it is by Fu It is represented by frequency domain after leaf transformationThen by the adjoint matrix X per section audio reference signal* k N audio residual signals E (n) in () and frequency domain can determine the corresponding renewal in time domain of every section of adaptive-filtering coefficient Factor lambdak(n), and λkN () is specifically represented by:λk(n)=IFFT (E (n) X* k(n)), by λkN the data of preceding L length are made in () It is effective updating factor, is represented by λ 'k(n), to each λ 'kN () carries out Fourier transformation, be finally available from adaptive filtering The updating factor Λ of each band frequency domain filter coefficient of devicek(n), and each band frequency domain filter coefficient updating factor Λk N () is represented by:
Wherein, k=0,1,2 ..., K-1.
S310, the adaptive-filtering coefficient is updated according to the updating factor and is preserved.
Connect above-mentioned example, it is known that the corresponding each band frequency domain filtering in frequency domain of the current adaptive-filtering coefficient for determining Device coefficient is represented by Wk (n), it is determined that the updating factor Λ of each band frequency domain filter coefficientkAfter (n), according to Wk (n)And Λk(n), it is possible to it is determined that the frequency domain filter coefficient W after updatingk (n+1), and Wk (n+1)Can be expressed as:
Wk (n+1)=Wk (n)+μΛk(n), wherein, the value of μ is typically set at the constant more than 0 and less than 1, and it specifically takes Value is preferable to be set according to historical experience value.
In the present embodiment, can be to each band frequency domain filter coefficient W of acquisitionk (n+1)Carry out inverse Fourier transform and It is cumulative, finally it is available for processing the adaptive-filtering coefficient of next frame audio input signal.
The present embodiment above-mentioned steps not only realize the treatment of audio input signal, also achieve adaptive-filtering coefficient Update, the present embodiment can be illustrated by a general principle figure to the acoustic signal processing method for being proposed.
Specifically, Fig. 4 b are that a kind of acoustic signal processing method provided based on the embodiment of the present invention three carries out audio signal The general principle figure for the treatment of, as shown in Figure 4 b, can be obtained the audio reference signal x of acquisition by sef-adapting filter first Echo estimates signalAfterwards can according to audio input signal d (in d may include through echo path w formed echo signal y, May also contain user voice signal v and noise signal n) and obtain echo and estimate signalCarry out echo cancellor and obtain right The audio residual signals e for answering;Then the noise signal in audio residual signals e can be eliminated according to noise signal cancellation module to obtain Audio enhancement signal e1 is obtained, and the residual noise in audio enhancement signal e1 is eliminated by residual noise cancellation module, finally Obtain exportable target audio signal S;Meanwhile, can be residual according to audio reference signal x, audio input signal d and audio Whether difference signal e includes in detecting currently processed audio input signal by the double-end signal detection module also illustrated in Fig. 4 b Whether user voice signal includes echo signal, determines whether to carry out adaptive-filtering coefficient more thus according to testing result Newly.
A kind of acoustic signal processing method that the embodiment of the present invention three is provided, not only realizes the place of audio input signal Reason, the renewal that also specifically increased adaptive-filtering coefficient judges and updates operation, interrelated between each operation so that adaptive The convergent ideal situation of evaluated error for answering wave filter to reach, more preferably improves echo and noise in audio interaction Eradicating efficacy, so that the elimination performance of the echo of audio interactive device and noise.
Example IV
Fig. 5 a are the schematic flow sheet of the preferred embodiment of the acoustic signal processing method that the embodiment of the present invention four is provided, this The application scenarios of inventive embodiments can be expressed as:One audio signal that foregoing invention embodiment can be applied to provide is provided first The audio interactive device (e.g., intelligent sound robot) of processing method, its audio input device is in intelligent sound robot Microphone, audio output apparatus are the loudspeaker in intelligent sound robot;It is then based on the above-mentioned Audio Signal Processing of this implementation The frame of method one carries out audio letter for unit to set input-buffer area sound intermediate frequency input signal in the intelligent sound robot Number treatment, wherein, the audio input signal in input-buffer area be specifically based on microphone pickup.
The present embodiment contains the audio input signal picked up under 3 kinds of scene modes, and 3 kinds of scene modes are respectively:Distal end Scene mode, this kind of pattern is not spoken for user, and only the audio output apparatus of intelligent audio interactive device carry out audio letter Number output;Echo signal and ambient noise are included in the audio input signal of now audio input device pickup;Both-end scene mould Formula, this kind of pattern is being spoken for user, and the audio output apparatus of intelligent audio interactive device are also carrying out audio signal output;This When audio input device pickup audio input signal in include user speech, echo signal and ambient noise;Near-end scene Pattern, this kind of pattern is user speaking, but the audio output apparatus of intelligent audio interactive device not carry out audio signal defeated Go out, user speech and ambient noise are included in the audio input signal of now audio input device pickup.It is understood that this Audio input signal and audio reference signal in embodiment are cached to corresponding input-buffer area and reference buffer storage area.
Fig. 5 b are the oscillogram of the preferred embodiment sound intermediate frequency input signal that the embodiment of the present invention four is provided, such as Fig. 5 b institutes Show, sets forth 3 kinds of oscillograms of the audio input signal of scene mode subaudio frequency input equipment pickup;Fig. 5 c are the present invention The oscillogram of the preferred embodiment sound intermediate frequency reference signal that example IV is provided, as shown in Figure 5 c, sets forth 3 kinds of scene moulds The oscillogram of the audio reference signal of formula subaudio frequency output equipment output.
As shown in Figure 5 a, based on the above embodiment of the present invention provide acoustic signal processing method to audio input signal at Reason, specifically includes following operation:
S401, obtain a frame audio reference signal from reference buffer storage area and to obtain from input-buffer area a frame audio defeated Enter signal, according to the currently adaptive-filtering coefficient processing of determination audio reference signal, obtain corresponding frame echo and estimate Signal.
Exemplarily, the reference buffer storage area is arranged in intelligent sound robot, the audio in the reference buffer storage area Reference signal is the signal of loudspeaker output to be based on.
S402, determine that the frame audio input signal and the frame echo estimate the signal difference of signal, using the signal difference as right The frame audio residual signals answered, subsequently perform S403 and S409 respectively.
S403, the power spectrum according to the frame audio residual signals and noise in the predetermined frame audio residual signals The noise power spectrum of signal, determines first posteriori SNR and prior weight of the frame audio residual signals, subsequently holds respectively Row S404 and S405;
S404, the update condition according to noise power spectrum, update the noise power spectrum and preserve, subsequent execution S411.
Exemplarily, the update condition of noise power spectrum can determine that noise estimates regulation parameter in the present embodiment, its In, the noise estimates that regulation parameter specifically can be true according to first posteriori SNR and prior weight of audio residual signals It is fixed, judge that whether noise estimates regulation parameter less than setting regulation parameter threshold value, if it is less, the noise power may be updated Spectrum, wherein, the value of the setting regulation parameter threshold value is typically set at the constant more than 0 and less than 1, and its specific value can be excellent That selects sets according to historical experience value, it is preferable that it is 0.4 to set the regulation parameter threshold value that sets.
S405, the first posteriori SNR and prior weight according to determination, determine that the frame audio residual signals are corresponding One frame audio enhancement signal.
S406, detect with the presence or absence of residual noise signal in the frame audio enhancement signal, if so, then performing S407;If it is not, Then perform S408.
S407, residual noise signal is filtered, and export a frame from audio output apparatus carries out residual noise filtering Audio enhancement signal afterwards, subsequent execution S411.
S408, export the frame audio enhancement signal, subsequent execution S411 from audio output apparatus.
S409, whether detect in the frame audio input signal comprising user voice signal, and detect that the frame audioref is believed Whether echo signal is included in number.
S410, when the user voice signal is included in the audio input signal, stop updating self adaptation filter Wave system number;Or, in the user voice signal and the audio reference signal is not included in the audio input signal not During comprising echo signal, stop updating the adaptive-filtering coefficient;Or, when in the audio input signal do not include institute When stating in user voice signal and the audio reference signal comprising echo signal, according to the frame audio reference signal and determine Audio residual signals, determine the updating factor of frequency domain filter coefficient, and the adaptive-filtering is updated according to the updating factor Coefficient is simultaneously preserved.
S411, when reference buffer storage area or input-buffer area be space-time, return perform S401.
It is understood that the acoustic signal processing method that the present embodiment is provided is the operating method that a circulation is performed, As long as audio input signal and audio reference signal can be obtained, it is possible to which the audio input signal to obtaining is processed.
It should be noted that this gives the corresponding treatment effect figure of different audio signals processing method, specifically Ground, the treatment effect figure processed audio input signal in the preferred embodiment that Fig. 5 d are provided for the embodiment of the present invention four, Fig. 5 e are the treatment effect figure processed audio input signal based on a kind of existing acoustic signal processing method, exemplary Ground, Fig. 5 e be based on existing open source projects be derived from webpage real-time Communication for Power (Web Real-Time Communication, WebRTC the acoustic signal processing method in) carries out the design sketch of signal transacting;Fig. 5 f are to be believed based on existing another audio The treatment effect figure that number processing method is processed audio input signal, exemplarily, Fig. 5 f are based on existing item of increasing income Acoustic signal processing method in mesh Speex carries out the design sketch of signal transacting.By the treatment to Fig. 5 d, Fig. 5 e and Fig. 5 f Effect is compared, again it can be seen that no matter with which kind of scene mode pickup audio input signal, the noise signal in Fig. 5 d is almost All eliminate, more noise signal is still remained in Fig. 5 e and Fig. 5 f, it is possible thereby to determine, provided based on the embodiment of the present invention Acoustic signal processing method treatment after corresponding noise signal be considerably less than based on the treatment of two kinds of existing methods after it is corresponding Noise signal, i.e. the corresponding process performance of acoustic signal processing method provided in an embodiment of the present invention is substantially better than existing method Process performance.
Additionally, the present embodiment also proposed echo return fading gain (Echo Return Loss Enhan-cement, ERLE signal transacting critical parameter), judge comparison is carried out for the result to acoustic signal processing method, wherein The ERLE is represented by:
Wherein, r (n) is that the n-th frame audio that audio input device is picked up is defeated Enter the echo signal in signal;E (n) is n-th frame audio input signal through the residual echo signal after Audio Signal Processing, E [r2 (n)] represent echo signal power spectrum, E [e2(n)] represent residual echo signal power spectrum, ERLE (n) represent n-th frame sound Frequency input signal returns to fading gain through the echo after Audio Signal Processing.
Specifically, it is existing based on acoustic signal processing method provided in an embodiment of the present invention and set forth above other two kinds After some acoustic signal processing methods carry out signal transacting, it is found that compared with other two kinds of existing methods set forth above, The embodiment of the present invention provides the corresponding echo return fading gain of acoustic signal processing method and can be higher by 10dB~15dB, by This also preferably embodies the process performance advantage of acoustic signal processing method provided in an embodiment of the present invention.
Embodiment five
Fig. 6 is a kind of structured flowchart of audio signal processing that the embodiment of the present invention five is provided, and the system is applied to Situation about being processed the audio signal being input into when audio is interacted, can be realized, and be typically integrated in by software and/or hardware On audio interactive device with audio interactive function.As shown in fig. 6, the audio signal processing includes:Echo signal disappears Except module 51, noise signal cancellation module 52, residual noise detection module 53 and residual noise filtering module 54.
Wherein, echo signal cancellation module 51, for carrying out echo to the echo signal in acquired audio input signal Eliminate, obtain the audio residual signals of the audio input signal.
Noise signal cancellation module 52, for carrying out noise elimination to the noise signal in the audio residual signals, obtains Obtain the audio enhancement signal after noise is eliminated.
Residual noise detection module 53, for carrying out residual noise signal detection to the audio enhancement signal.
Residual noise filtering module 54, for when there is residual noise signal in the audio enhancement signal, to described Residual noise signal is filtered.
In the present embodiment, the audio signal processing passes through audio acquired in echo signal cancellation module 51 pairs first Echo signal in input signal carries out echo cancellor, obtains the audio residual signals of the audio input signal;Then pass through Noise signal in the audio residual signals of noise signal cancellation module 52 pairs carries out noise elimination, after obtaining noise elimination Audio enhancement signal;Afterwards residual noise signal inspection is carried out by 53 pairs of audio enhancement signals of residual noise detection module Survey;Eventually through residual noise filtering module 54 when there is residual noise signal in the audio enhancement signal, to described residual Noise signal is stayed to be filtered.
A kind of audio signal processing that the embodiment of the present invention five is provided, effectively eliminates audio input when audio is interacted Echo signal and ambient noise signal in signal, while detection and filtering to residual noise are realized, with existing method phase Than, influence of the ambient noise to audio interaction is effectively inhibited, more preferably improve echo and noise in audio interaction Eradicating efficacy, so as to improve audio interactive device to echo and the elimination performance of noise.
Further, noise signal cancellation module 52, including:
First signal to noise ratio determining unit 521, for the power spectrum according to the audio residual signals and predetermined The noise power spectrum of noise signal in the audio residual signals, determine the audio residual signals the first posteriori SNR and Prior weight.
First attenuation coefficient determining unit 522, for according to first posteriori SNR and the prior weight, really First attenuation coefficient of noise signal in the fixed audio residual signals.
Enhancing signal acquiring unit 523, for according in first attenuation coefficient elimination audio residual signals Noise signal, obtains the audio enhancement signal of the audio residual signals.
On the basis of above-mentioned optimization, noise signal cancellation module 52 also includes:
Regulation parameter determining unit 524, for it is determined that first posteriori SNR and priori of the audio residual signals After signal to noise ratio, according to first posteriori SNR and the prior weight, the noise of the audio residual signals is determined Estimate regulation parameter.
Noise power spectrum updating block 525, for estimating regulation parameter less than setting regulation parameter threshold value when the noise When, update the noise power spectrum and preserve.
Further, the first signal to noise ratio determining unit 521, specifically for respectively according to following first posteriori SNR formula And prior weight formula obtains first posteriori SNR and prior weight of the audio residual signals, after described first Signal to noise ratio formula is tested to be expressed as:
The prior weight formula is expressed as:
Wherein, j represents the frequency number of frequency domain, is any integer value in 0~M-1, and M represents frequency length of field, n tables Show the frame number for being presently in managing audio input signal, α is the either constant between 0~1, SNRpost(n, j) represents n-th frame audio First posteriori SNR of the audio residual signals of input signal, SNRprio(n, j) represents the audio of n-th frame audio input signal The prior weight of residual signals, | E (n, j) |2The power spectrum of the audio residual signals of n-th frame audio input signal is represented, | NE (n,j)|2The noise power spectrum of noise signal in the audio residual signals of n-th frame audio input signal is represented, | E1 (n-1, j) |2 Represent the enhancing power spectrum signal of the audio residual signals of the (n-1)th frame audio input signal.
First attenuation coefficient determining unit 522, specifically for obtaining the audio according to following first attenuation coefficient formula First attenuation coefficient of noise signal in residual signals, the first attenuation coefficient formula is expressed as:
Wherein, ζ 1 (n, j) represents the first attenuation coefficient of the audio residual signals of n-th frame audio input signal.
Accordingly, strengthen signal acquiring unit 523, be additionally operable to determine n-th frame according to following enhancing power spectrum signal formula The enhancing power spectrum signal of the audio residual signals of audio input signal, the enhancing power spectrum signal formula is expressed as:
|E1(n,j)|2=ζ 1 (n, j) | E (n, j) |2,
Wherein, | E1 (n, j) |2Represent the enhancing power spectrum signal of the audio residual signals of n-th frame audio input signal.
Regulation parameter determining unit 524, specifically for estimating that regulation parameter formula obtains the audio according to following noises The noise of residual signals estimates regulation parameter, and the noise estimates that regulation parameter formula is expressed as:
Wherein, β represents that the noise of n-th frame audio residual signals estimates regulation parameter;
Noise power spectrum updating block 525, specifically for updating the noise power according to following power spectrum more new formula Spectrum, the power spectrum more new formula is expressed as:
|NE(n,j)|2=ε | NE (n-1, j) |2+(1-ε)|E(n,j)|2,
Wherein, | NE (n-1, j) |2Represent the noise of noise signal in the audio residual signals of n-th frame audio input signal Power spectrum.
Further, residual noise detection module 53, including:
Detection parameter determination unit 531, for according to the audio input signal, the audio enhancement signal and default Detection parameter equation, determines the residual noise detection parameter of the audio enhancement signal;
Residual noise determining unit 532, for when detection parameter threshold of the residual noise detection parameter less than setting When, determine there is residual noise signal in the audio enhancement signal;
Second attenuation coefficient determining unit 533, for detecting parameter and the detection parameter according to the residual noise Threshold value, obtains second attenuation coefficient of the residual noise signal in the audio enhancement signal.
Accordingly, residual noise filtering module 54, including:
Residual noise removal unit 541, for according to second attenuation coefficient, filtering the audio enhancement signal in Residual noise signal.
On the basis of above-mentioned optimization, the detection parameter equation is expressed as:
Wherein, n represents the frame number for being presently in managing audio input signal, ∑ | d (n, t) |2Represent the energy value of n-th frame audio input signal;∑|e1(n,t)|2Represent the audio enhancing letter of n-th frame audio input signal Number energy value;RD (n) represents the audio enhancement signal residual noise detection parameter of n-th frame audio input signal.
Second attenuation coefficient determining unit 533, specifically for determining the residual according to following second attenuation coefficient formula Second attenuation coefficient of the noise signal in the audio enhancement signal, the second attenuation coefficient formula is expressed as:ζ 2 (n)= max(RDth/RD(n),ζmin), wherein, RDthIt is the detection parameter threshold of setting;ζminIt is the minimal attenuation coefficient of setting;ζ2 Second attenuation coefficient of residual noise signal in the audio enhancement signal of (n) expression n-th frame audio input signal.
Further, residual noise filtering module 54, also includes:
Echo signal determining unit 542, for after being filtered to the residual noise signal, according to following signals Determine formula, determine target audio signal to be output, the signal determines that formula is expressed as:S (n)=ζ 2 (n) * e1 (n), its In, e1 (n) represents the audio enhancement signal of n-th frame, and s (n) represents n-th frame target audio signal to be output.
Further, echo signal cancellation module 51, including:
Audio signal acquiring unit 511, the audio input signal and audio for obtaining audio input device pickup is defeated Go out the audio reference signal of equipment output.
Signal determining unit 512 is estimated, for the audioref according to the current adaptive-filtering coefficient processing for determining Signal, the echo for obtaining the audio reference signal estimates signal.
Residual signals determining unit 513, for the audio input signal and the echo to be estimated into the difference of signal is true It is set to the audio residual signals of the audio input signal.
Additionally, echo signal cancellation module 51, also includes:
Updating factor determining unit 514, for the audio input signal to be estimated the difference of signal with the echo It is defined as after the audio residual signals of the audio input signal, according to the audio reference signal and the audio residual error for determining Signal, determines the updating factor of frequency domain filter coefficient.
Filter factor processing unit 515, for updating the adaptive-filtering coefficient according to the updating factor and preserving.
On the basis of above-mentioned optimization, the audio signal processing also includes:
Double-end signal detection module 55, carries out echo and disappears for the echo signal in acquired audio input signal Remove, obtain after the audio residual signals of the audio input signal, whether user is included in the detection audio input signal Voice signal, and whether detect in the audio reference signal comprising echo signal.
Accordingly, filter factor processing unit 515, is additionally operable in the audio input signal comprising the user speech During signal, stop updating the adaptive-filtering coefficient;Or, when in the audio input signal do not include the user speech When not including echo signal in signal and the audio reference signal, stop updating the adaptive-filtering coefficient;Or, work as institute When stating in audio input signal not comprising in the user voice signal and the audio reference signal comprising echo signal, update The adaptive-filtering coefficient is simultaneously preserved.
Further, double-end signal detection module 55, including:
Energy ratio determining unit 551, for the energy value according to the audio input signal and the audio reference signal Energy value, determine the instantaneous energy ratio of the audio input signal and the audio reference signal.
Voice signal identifying unit 552, for when the instantaneous energy ratio is less than the first given threshold, determining the sound User voice signal is not included in frequency input signal;Otherwise, it determines comprising the user voice signal.
Additionally, double-end signal detection module 55, also includes:
Second signal to noise ratio determining unit 553, for the power spectrum according to the audio input signal and the audio residual error The corresponding noise power spectrum of signal, determines the second posteriori SNR of the audio input signal;
Cross-correlation coefficient determining unit 554, for the performance number according to the audio input signal and the audio residual error The performance number of signal, determines the cross-correlation coefficient of the audio input signal and the audio residual signals;
Echo signal identifying unit 555, for when second posteriori SNR is less than the second given threshold, determining institute State in audio reference signal not comprising echo signal;Or, when second posteriori SNR not less than the second given threshold and When the cross-correlation coefficient is not less than three given thresholds, it is determined that comprising echo signal and described in the audio reference signal The echo path of echo signal there occurs change;Or, when second posteriori SNR is not less than second given threshold And the cross-correlation coefficient is when being less than three given threshold, determine in the audio reference signal comprising echo signal and institute The echo path for stating echo signal does not change.
The specific work process of the present embodiment sound intermediate frequency signal processing system may be referred to embodiment one to example IV, This is repeated no more.
Embodiment six
The embodiment of the present invention six provides a kind of audio interactive device, and it is above-mentioned that the audio interactive device includes the present invention A kind of audio signal processing that embodiment is provided.
Further, also audio interactive device further comprises audio input device, for picking up audio input signal;Audio Output equipment, for exporting audio reference signal.
In the present embodiment, can be by audio signal processing, according to audio input based on the audio interactive device The audio input signal of equipment pickup and the audio reference signal of audio output apparatus output carry out Audio Signal Processing, so as to disappear Except echo signal and noise signal in audio input signal, while can also enter to residual noise signal after noise signal is eliminated Row detection, compared with existing audio interactive device, effectively inhibits influence of the ambient noise to audio interaction, more preferably carries The eradicating efficacy of echo and noise in audio interaction is risen, echo and noise have been disappeared so as to improve audio interactive device Except performance.
The embodiment of the present invention additionally provides a kind of computer equipment, and the computer equipment can be integrated as audio interactive device The audio signal processing that above-described embodiment is provided.Fig. 7 is a kind of knot of computer equipment that the embodiment of the present invention six is provided Structure block diagram, as shown in fig. 7, the computer equipment also includes:Processor 61 and storage device 62;The quantity of processor 61 in equipment Can be one or more, in Fig. 7 by taking a processor 61 as an example;Processor 61 and storage device 62 in the equipment can be with Connected by bus or other modes, in Fig. 7 as a example by being connected by bus.
Storage device 62 can be used to store one or more programs, the journey as a kind of computer-readable recording medium Sequence can be software program, computer executable program and module, such as acoustic signal processing method in the embodiment of the present invention Corresponding programmed instruction/module is (for example, echo signal cancellation module 51 in audio signal processing shown in accompanying drawing 6, make an uproar Acoustical signal cancellation module 52, residual noise detection module 53 and residual noise filtering module 54).Processor 61 is deposited by operation Store up the software program in storage device 62, instruct and module, so as to perform at the various function application and data of equipment Reason, that is, realize the acoustic signal processing method in above method embodiment.
Storage device 62 may include storing program area and storage data field, wherein, storing program area can storage program area, Application program needed at least one function;Storage data field can be stored and use created data etc. according to equipment.Additionally, Storage device 62 can include high-speed random access memory, can also include nonvolatile memory, for example, at least one magnetic Disk storage device, flush memory device or other non-volatile solid state memory parts.In some instances, storage device 62 can enter one Step includes the memory remotely located relative to processor 61, and these remote memories can be by network connection to equipment.On State example including but not limited to internet, intranet, LAN, mobile radio communication and combinations thereof of network.
Also, when one or more program included by above computer equipment is by one or more of processors 61 During execution, program is proceeded as follows:
Echo cancellor is carried out to the echo signal in acquired audio input signal, the sound of the audio input signal is obtained Frequency residual signals;Noise elimination is carried out to the noise signal in the audio residual signals, the audio after noise is eliminated is obtained and is increased Strong signal;Residual noise signal detection is carried out to the audio enhancement signal, if there is residual in the audio enhancement signal Noise signal, then filter to the residual noise signal.
Additionally, the embodiment of the present invention provides a kind of storage medium comprising computer executable instructions, the computer again Executable instruction is used to perform a kind of such as embodiment one, embodiment two, embodiment three or real when being performed by computer processor The acoustic signal processing method described in example four is applied, the method includes:Echo signal in acquired audio input signal is entered Row echo cancellor, obtains the audio residual signals of the audio input signal;To the noise signal in the audio residual signals Noise elimination is carried out, the audio enhancement signal after noise is eliminated is obtained;Residual noise signal is carried out to the audio enhancement signal Detection, if there is residual noise signal in the audio enhancement signal, filters to the residual noise signal.
By the description above with respect to implementation method, it is apparent to those skilled in the art that, the present invention Can be realized by software and required common hardware, naturally it is also possible to realized by hardware, but the former is more in many cases Good implementation method.Based on such understanding, what technical scheme substantially contributed to prior art in other words Part can be embodied in the form of software product, and the computer software product can be stored in computer-readable recording medium In, such as floppy disk, read-only storage (Read-Only Memory, ROM), the random access memory (Random of computer Access Memory, RAM), flash memory (FLASH), hard disk or CD etc., including some instructions are used to so that a computer sets Standby (can be personal computer, server, or network equipment etc.) performs the method described in each embodiment of the invention.
It is worth noting that, in the embodiment of above-mentioned audio signal processing, included unit and module are only Divided according to function logic, but be not limited to above-mentioned division, as long as corresponding function can be realized;Separately Outward, the specific name of each functional unit is also only to facilitate mutually distinguish, the protection domain being not intended to limit the invention.
Note, above are only presently preferred embodiments of the present invention and institute's application technology principle.It will be appreciated by those skilled in the art that The invention is not restricted to specific embodiment described here, can carry out for a person skilled in the art various obvious changes, Readjust and substitute without departing from protection scope of the present invention.Therefore, although the present invention is carried out by above example It is described in further detail, but the present invention is not limited only to above example, without departing from the inventive concept, also More other Equivalent embodiments can be included, and the scope of the present invention is determined by scope of the appended claims.

Claims (28)

1. a kind of acoustic signal processing method, it is characterised in that including:
Echo cancellor is carried out to the echo signal in acquired audio input signal, the audio for obtaining the audio input signal is residual Difference signal;
Noise elimination is carried out to the noise signal in the audio residual signals, the audio enhancement signal after noise is eliminated is obtained;
Residual noise signal detection is carried out to the audio enhancement signal, if there is residual noise in the audio enhancement signal Signal, then filter to the residual noise signal.
2. method according to claim 1, it is characterised in that the noise signal in the audio residual signals is entered Row noise is eliminated, and obtains the audio enhancement signal after noise is eliminated, including:
Noise signal makes an uproar in power spectrum and the predetermined audio residual signals according to the audio residual signals Power sound spectrum, determines first posteriori SNR and prior weight of the audio residual signals;
According to first posteriori SNR and the prior weight, of noise signal in the audio residual signals is determined One attenuation coefficient;
Noise signal in the audio residual signals is eliminated according to first attenuation coefficient, the audio residual signals are obtained Audio enhancement signal.
3. method according to claim 2, it is characterised in that in the power spectrum according to the audio residual signals and pre- The noise power spectrum of noise signal, determines the first posteriority of the audio residual signals in the audio residual signals for first determining After signal to noise ratio and prior weight, also include:
According to first posteriori SNR and the prior weight, determine that the noise of the audio residual signals estimates regulation Parameter;
If the noise estimates that regulation parameter, less than setting regulation parameter threshold value, updates the noise power spectrum and preserves.
4. method according to claim 3, it is characterised in that respectively according to following first posteriori SNR formula and elder generation The first posteriori SNR and prior weight that signal to noise ratio formula obtains the audio residual signals are tested,
The first posteriori SNR formula is expressed as:
The prior weight formula is expressed as:
SNR p r i o ( n , j ) = &alpha; | E 1 ( n - 1 , j ) | 2 | N E ( n , j ) | 2 + ( 1 - &alpha; ) m a x ( SNR p o s t ( n , j ) - 1 , 0 ) ,
Wherein, j represents the frequency number of frequency domain, is any integer value in 0~M-1, and M represents frequency length of field, and n is represented and worked as The frame number of preceding handled audio input signal, α is the either constant between 0~1, SNRpost(n, j) represents n-th frame audio input First posteriori SNR of the audio residual signals of signal, SNRprio(n, j) represents the audio residual error of n-th frame audio input signal The prior weight of signal, | E (n, j) |2The power spectrum of the audio residual signals of n-th frame audio input signal is represented, | NE (n, j)|2The noise power spectrum of noise signal in the audio residual signals of n-th frame audio input signal is represented, | E1 (n-1, j) |2Represent The enhancing power spectrum signal of the audio residual signals of the (n-1)th frame audio input signal;
The first attenuation coefficient of noise signal in the audio residual signals is obtained according to following first attenuation coefficient formula, it is described First attenuation coefficient formula is expressed as:
&zeta; 1 ( n , j ) = SNR p r i o ( n , j ) 1 + SNR p r i o ( n , j ) exp ( 1 2 exp int ( SNR p r i o ( n , j ) SNR p o s t ( n , j ) 1 + SNR p r i o ( n , j ) ) ) ,
Wherein, ζ 1 (n, j) represents the first attenuation coefficient of the audio residual signals of n-th frame audio input signal;
Accordingly, the increasing of the audio residual signals of n-th frame audio input signal is determined according to following enhancing power spectrum signal formula Strong power spectrum signal, the enhancing power spectrum signal formula is expressed as:
|E1(n,j)|2=ζ 1 (n, j) | E (n, j) |2,
Wherein, | E1 (n, j) |2Represent the enhancing power spectrum signal of the audio residual signals of n-th frame audio input signal;
Estimate that regulation parameter formula obtains the noise estimation regulation parameter of the audio residual signals according to following noises, it is described to make an uproar Sound estimates that regulation parameter formula is expressed as:
&beta; = 1 M &Sigma; j = 0 M - 1 ( SNR p r i o ( n , j ) SNR p o s t ( n , j ) 1 + SNR p r i o ( n , j ) - log ( 1 + SNR p r i o ( n , j ) ) ) ,
Wherein, β represents that the noise of n-th frame audio residual signals estimates regulation parameter;
The noise power spectrum is updated according to following noise power spectrums more new formula, the noise power spectrum more new formula is represented For:|NE(n,j)|2=ε | NE (n-1, j) |2+(1-ε)|E(n,j)|2,
Wherein, | NE (n-1, j) |2Represent the noise power of noise signal in the audio residual signals of the (n-1)th frame audio input signal Spectrum.
5. method according to claim 1, it is characterised in that residual noise signal inspection is carried out to the audio enhancement signal Survey, if there is residual noise signal in the audio enhancement signal, the residual noise signal is filtered, including:
According to the audio input signal, the audio enhancement signal and default detection parameter equation, determine that the audio increases The residual noise detection parameter of strong signal;
If the residual noise detection parameter is less than the detection parameter threshold of setting, it is determined that deposited in the audio enhancement signal In residual noise signal;
Parameter and the detection parameter threshold are detected according to the residual noise, the residual noise signal is obtained in the sound The second attenuation coefficient in frequency enhancing signal;
According to second attenuation coefficient, the residual noise signal in the audio enhancement signal is filtered.
6. method according to claim 5, it is characterised in that the detection parameter equation is expressed as:
Wherein, n represents the frame number for being presently in managing audio input signal, ∑ | d (n, t) |2Represent The energy value of n-th frame audio input signal;∑|e1(n,t)|2Represent the energy of the audio enhancement signal of n-th frame audio input signal Value;RD (n) represents the residual noise detection parameter of the audio enhancement signal of n-th frame audio input signal;
Determine the residual noise signal second declining in the audio enhancement signal according to following second attenuation coefficient formula Subtract coefficient,
The second attenuation coefficient formula is expressed as:ζ 2 (n)=max (RDth/RD(n),ζmin), wherein, RDthIt is the detection of setting Parameter threshold;ζminIt is the minimal attenuation coefficient of setting;It is residual in the audio enhancement signal of ζ 2 (n) expression n-th frame audio input signals Stay the second attenuation coefficient of noise signal.
7. the method according to right wants 6, it is characterised in that after being filtered to the residual noise signal, also wrap Include:
Determine formula according to following signals, determine target audio signal to be output;
The signal determines that formula is expressed as:S (n)=ζ 2 (n) * e1 (n), wherein, e1 (n) represents n-th frame audio input signal Audio enhancement signal, s (n) represents the target audio signal of n-th frame audio input signal to be output.
8. method according to claim 1, it is characterised in that the echo signal in acquired audio input signal Echo cancellor is carried out, the audio residual signals of the audio input signal are obtained, including:
Obtain the audio input signal of audio input device pickup and the audio reference signal of audio output apparatus output;
The audio reference signal according to the current adaptive-filtering coefficient processing for determining, obtains returning for the audio reference signal Sound estimates signal;
The audio that the difference of the audio input signal and echo estimation signal is defined as into the audio input signal is residual Difference signal.
9. method according to claim 8, it is characterised in that the audio input signal and the echo are being estimated into letter Number difference be defined as after the audio residual signals of the audio input signal, also include:
According to the audio reference signal and the audio residual signals for determining, the updating factor of frequency domain filter coefficient is determined;
The adaptive-filtering coefficient is updated according to the updating factor and is preserved.
10. method according to claim 9, it is characterised in that the echo signal in acquired audio input signal Echo cancellor is carried out, is obtained after the audio residual signals of the audio input signal, also included:
Whether detect in the audio input signal comprising user voice signal, and detect and whether wrap in the audio reference signal Containing echo signal;
When the user voice signal is included in the audio input signal, stop updating the adaptive-filtering coefficient;Or Person, not comprising echo letter in the user voice signal and the audio reference signal is not included in the audio input signal Number when, stop updating the adaptive-filtering coefficient;Or, when in the audio input signal not comprising the user speech believe Number and the audio reference signal in comprising echo signal when, update the adaptive-filtering coefficient and simultaneously preserve.
11. methods according to claim 10, it is characterised in that whether included in the detection audio input signal User voice signal, including:
The energy value of energy value and the audio reference signal according to the audio input signal, determines the audio input letter Instantaneous energy ratio number with the audio reference signal;
If the instantaneous energy ratio is less than the first given threshold, it is determined that do not include user speech in the audio input signal Signal;Otherwise, it determines comprising the user voice signal.
12. methods according to claim 10, it is characterised in that whether included in the detection audio reference signal Echo signal, including:
Power spectrum and the corresponding noise power spectrum of the audio residual signals according to the audio input signal, determine the sound Second posteriori SNR of frequency input signal;
The performance number of performance number and the audio residual signals according to the audio input signal, determines the audio input letter Cross-correlation coefficient number with the audio residual signals;
If second posteriori SNR is less than the second given threshold, it is determined that do not include echo in the audio reference signal Signal;Or, if second posteriori SNR is not less than the second given threshold and the cross-correlation coefficient is not less than the 3rd Given threshold, it is determined that the echo path comprising echo signal and the echo signal in the audio reference signal there occurs change Change;Or, if second posteriori SNR is not less than second given threshold and the cross-correlation coefficient is less than described 3rd given threshold, it is determined that the echo path comprising echo signal and the echo signal in the audio reference signal does not have Change.
A kind of 13. audio signal processings, it is characterised in that including:
Echo signal cancellation module, for carrying out echo cancellor to the echo signal in acquired audio input signal, obtains institute State the audio residual signals of audio input signal;
Noise signal cancellation module, for carrying out noise elimination to the noise signal in the audio residual signals, obtains noise Audio enhancement signal after elimination;
Residual noise detection module, for carrying out residual noise signal detection to the audio enhancement signal;
Residual noise filtering module, for when there is residual noise signal in the audio enhancement signal, being made an uproar to the residual Acoustical signal is filtered.
14. systems according to claim 13, it is characterised in that the noise signal cancellation module, including:
First signal to noise ratio determining unit, for the power spectrum according to the audio residual signals and the predetermined audio The noise power spectrum of noise signal in residual signals, determines first posteriori SNR and priori noise of the audio residual signals Than;
First attenuation coefficient determining unit, for according to first posteriori SNR and the prior weight, it is determined that described First attenuation coefficient of noise signal in audio residual signals;
Enhancing signal acquiring unit, for eliminating the noise letter in the audio residual signals according to first attenuation coefficient Number, obtain the audio enhancement signal of the audio residual signals.
15. systems according to claim 14, it is characterised in that the noise signal cancellation module, also include:
Regulation parameter determining unit, for it is determined that the audio residual signals the first posteriori SNR and prior weight it Afterwards, according to first posteriori SNR and the prior weight, determine that the noise of the audio residual signals estimates regulation Parameter;
Noise power spectrum updating block, for when the noise estimates regulation parameter less than setting regulation parameter threshold value, updating The noise power spectrum is simultaneously preserved.
16. systems according to claim 15, it is characterised in that the first signal to noise ratio determining unit, specifically for dividing The first posteriority of the audio residual signals is not obtained according to following first posteriori SNR formula and prior weight formula Signal to noise ratio and prior weight,
The first posteriori SNR formula is expressed as:
The prior weight formula is expressed as:
SNR p r i o ( n , j ) = &alpha; | E 1 ( n - 1 , j ) | 2 | N E ( n , j ) | 2 + ( 1 - &alpha; ) max ( SNR p o s t ( n , j ) - 1 , 0 ) ,
Wherein, j represents the frequency number of frequency domain, is any integer value in 0~M-1, and M represents frequency length of field, and n is represented and worked as The frame number of preceding handled audio input signal, α is the either constant between 0~1, SNRpost(n, j) represents n-th frame audio input First posteriori SNR of the audio residual signals of signal, SNRprio(n, j) represents the audio residual error of n-th frame audio input signal The prior weight of signal, | E (n, j) |2The power spectrum of the audio residual signals of n-th frame audio input signal is represented, | NE (n, j)|2The noise power spectrum of noise signal in the audio residual signals of n-th frame audio input signal is represented, | E1 (n-1, j) |2Represent The enhancing power spectrum signal of the audio residual signals of the (n-1)th frame audio input signal;
The first attenuation coefficient determining unit, specifically for obtaining the audio residual error according to following first attenuation coefficient formula First attenuation coefficient of noise signal in signal,
The first attenuation coefficient formula is expressed as:
&zeta; 1 ( n , j ) = SNR p r i o ( n , j ) 1 + SNR p r i o ( n , j ) exp ( 1 2 exp int ( SNR p r i o ( n , j ) SNR p o s t ( n , j ) 1 + SNR p r i o ( n , j ) ) ) ,
Wherein, ζ 1 (n, j) represents the first attenuation coefficient of the audio residual signals of n-th frame audio input signal;
Accordingly, strengthen signal acquiring unit, be additionally operable to determine n-th frame audio input according to following enhancing power spectrum signal formula The enhancing power spectrum signal of the audio residual signals of signal, the enhancing power spectrum signal formula is expressed as:
|E1(n,j)|2=ζ 1 (n, j) | E (n, j) |2,
Wherein, | E1 (n, j) |2Represent the enhancing power spectrum signal of the audio residual signals of n-th frame audio input signal;
The regulation parameter determining unit, specifically for estimating that regulation parameter formula obtains the audio residual error according to following noises The noise of signal estimates regulation parameter,
The noise estimates that regulation parameter formula is expressed as:
&beta; = 1 M &Sigma; j = 0 M - 1 ( SNR p r i o ( n , j ) SNR p o s t ( n , j ) 1 + SNR p r i o ( n , j ) - l o g ( 1 + SNR p r i o ( n , j ) ) ) ,
Wherein, β represents that the noise of n-th frame audio residual signals estimates regulation parameter;
Noise power spectrum updating block, specifically for updating the noise power spectrum according to following noise power spectrums more new formula, The noise power spectrum more new formula is expressed as:
|NE(n,j)|2=ε | NE (n-1, j) |2+(1-ε)|E(n,j)|2,
Wherein, | NE (n-1, j) |2Represent the noise power of noise signal in the audio residual signals of the (n-1)th frame audio input signal Spectrum.
17. systems according to claim 13, it is characterised in that the residual noise detection module, including:
Detection parameter determination unit, for being joined according to the audio input signal, the audio enhancement signal and default detection Number formula, determines the residual noise detection parameter of the audio enhancement signal;
Residual noise determining unit, for when the residual noise detection parameter less than setting detection parameter threshold when, it is determined that There is residual noise signal in the audio enhancement signal;
Second attenuation coefficient determining unit, for detecting parameter and the detection parameter threshold according to the residual noise, obtains Obtain second attenuation coefficient of the residual noise signal in the audio enhancement signal;
Accordingly, the residual noise filtering module, including:
Residual noise removal unit, for according to second attenuation coefficient, the residual filtered in the audio enhancement signal to be made an uproar Acoustical signal.
18. systems according to claim 17, it is characterised in that the detection parameter equation is expressed as:
Wherein, n represents the frame number for being presently in managing audio input signal, ∑ | d (n, t) |2Represent The energy value of n-th frame audio input signal;∑|e1(n,t)|2Represent the energy of the audio enhancement signal of n-th frame audio input signal Value;RD (n) represents the residual noise detection parameter of the audio enhancement signal of n-th frame audio input signal;
Second attenuation coefficient determining unit, specifically for determining the residual noise signal according to following second attenuation coefficient formula The second attenuation coefficient in the audio enhancement signal,
The second attenuation coefficient formula is expressed as:ζ 2 (n)=max (RDth/RD(n),ζmin), wherein, RDthIt is the detection of setting Parameter threshold;ζminIt is the minimal attenuation coefficient of setting;It is residual in the audio enhancement signal of ζ 2 (n) expression n-th frame audio input signals Stay the second attenuation coefficient of noise signal.
19. systems according to claim 18, it is characterised in that the residual noise filtering module, also include:
Echo signal determining unit, for after being filtered to the residual noise signal, being determined according to following signals public Formula, determines target audio signal to be output;
The signal determines that formula is expressed as:S (n)=ζ 2 (n) * e1 (n), wherein, e1 (n) represents the audio enhancing letter of n-th frame Number, s (n) represents n-th frame target audio signal to be output.
20. systems according to claim 13, it is characterised in that the echo signal cancellation module, including:
Audio signal acquiring unit, the audio input signal and audio output apparatus for obtaining audio input device pickup is defeated The audio reference signal for going out;
Estimate signal determining unit, for the audio reference signal according to the current adaptive-filtering coefficient processing for determining, obtain The echo for obtaining the audio reference signal estimates signal;
Residual signals determining unit, it is described for the audio input signal and the echo to be estimated into the difference of signal is defined as The audio residual signals of audio input signal.
21. systems according to claim 20, it is characterised in that the echo signal cancellation module, also include:
Updating factor determining unit, for the audio input signal and the echo to be estimated into the difference of signal is defined as institute After stating the audio residual signals of audio input signal, according to the audio reference signal and the audio residual signals for determining, really Determine the updating factor of frequency domain filter coefficient;
Filter factor processing unit, for updating the adaptive-filtering coefficient according to the updating factor and preserving.
22. systems according to claim 20, it is characterised in that also include:
Double-end signal detection module, echo cancellor is carried out for the echo signal in acquired audio input signal, is obtained After the audio residual signals of the audio input signal, whether comprising user speech letter in the detection audio input signal Number, and whether detect in the audio reference signal comprising echo signal;
Accordingly, the filter factor processing unit, is additionally operable in the audio input signal comprising user speech letter Number when, stop updating the adaptive-filtering coefficient;Or, when in the audio input signal not comprising the user speech believe Number and the audio reference signal in not comprising echo signal when, stop updating the adaptive-filtering coefficient;Or, when described When in audio input signal not comprising in the user voice signal and the audio reference signal comprising echo signal, institute is updated State adaptive-filtering coefficient and preserve.
23. systems according to claim 22, it is characterised in that the double-end signal detection module, including:
Energy ratio determining unit, for the energy value according to the audio input signal and the energy of the audio reference signal Value, determines the instantaneous energy ratio of the audio input signal and the audio reference signal;
Voice signal identifying unit, for when the instantaneous energy ratio is less than the first given threshold, determining the audio input User voice signal is not included in signal;Otherwise, it determines comprising the user voice signal.
24. systems according to claim 22, it is characterised in that the double-end signal detection module, also include:
Second signal to noise ratio determining unit, for the power spectrum according to the audio input signal and audio residual signals correspondence Noise power spectrum, determine the second posteriori SNR of the audio input signal;
Cross-correlation coefficient determining unit, for the performance number according to the audio input signal and the work(of the audio residual signals Rate value, determines the cross-correlation coefficient of the audio input signal and the audio residual signals;
Echo signal identifying unit, for when second posteriori SNR is less than the second given threshold, determining the audio Echo signal is not included in reference signal;Or, when second posteriori SNR not less than the second given threshold and it is described mutually When coefficient correlation is not less than three given thresholds, it is determined that comprising echo signal and echo letter in the audio reference signal Number echo path there occurs change;Or, when second posteriori SNR is not less than second given threshold and described When cross-correlation coefficient is less than three given threshold, determine in the audio reference signal comprising echo signal and the echo The echo path of signal does not change.
25. a kind of audio interactive devices, it is characterised in that the audio interactive device includes any one of claim 13-24 institute The audio signal processing stated.
26. devices according to claim 25, it is characterised in that also include:Audio input device, it is defeated for picking up audio Enter signal;Audio output apparatus, for exporting audio reference signal.
27. a kind of computer equipments, it is characterised in that the computer equipment includes:
One or more processors;
Storage device, for storing one or more programs,
One or more of programs are by one or more of computing devices so that one or more of processors are realized Acoustic signal processing method as any one of claim 1-12.
A kind of 28. storage mediums comprising computer executable instructions, the computer executable instructions are by computer disposal For performing the acoustic signal processing method as any one of claim 1-12 when device is performed.
CN201710182290.XA 2017-03-24 2017-03-24 Audio signal processing method and system, audio interaction device and computer equipment Active CN106898359B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710182290.XA CN106898359B (en) 2017-03-24 2017-03-24 Audio signal processing method and system, audio interaction device and computer equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710182290.XA CN106898359B (en) 2017-03-24 2017-03-24 Audio signal processing method and system, audio interaction device and computer equipment

Publications (2)

Publication Number Publication Date
CN106898359A true CN106898359A (en) 2017-06-27
CN106898359B CN106898359B (en) 2020-03-17

Family

ID=59193462

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710182290.XA Active CN106898359B (en) 2017-03-24 2017-03-24 Audio signal processing method and system, audio interaction device and computer equipment

Country Status (1)

Country Link
CN (1) CN106898359B (en)

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107393550A (en) * 2017-07-14 2017-11-24 深圳永顺智信息科技有限公司 Method of speech processing and device
CN107393546A (en) * 2017-09-04 2017-11-24 恒玄科技(上海)有限公司 A kind of echo cancel method and speech recognition apparatus for speech recognition process
CN108696648A (en) * 2018-05-16 2018-10-23 北京小鱼在家科技有限公司 A kind of method, apparatus, equipment and the storage medium of Short Time Speech signal processing
CN109256145A (en) * 2017-07-14 2019-01-22 北京搜狗科技发展有限公司 Audio-frequency processing method, device, terminal and readable storage medium storing program for executing based on terminal
CN109658946A (en) * 2017-10-12 2019-04-19 深圳前海黑鲸科技有限公司 A kind of echo processing method, device, storage medium and terminal device
CN109785853A (en) * 2019-03-11 2019-05-21 出门问问信息科技有限公司 A kind of echo cancel method, device, system and storage medium
CN110021289A (en) * 2019-03-28 2019-07-16 腾讯科技(深圳)有限公司 A kind of audio signal processing method, device and storage medium
CN110176244A (en) * 2018-06-19 2019-08-27 腾讯科技(深圳)有限公司 Echo cancel method, device, storage medium and computer equipment
CN110418233A (en) * 2019-07-26 2019-11-05 歌尔股份有限公司 A kind of earphone noise-reduction method, device, earphone and readable storage medium storing program for executing
CN111048118A (en) * 2019-12-24 2020-04-21 大众问问(北京)信息科技有限公司 Voice signal processing method and device and terminal
CN111402910A (en) * 2018-12-17 2020-07-10 华为技术有限公司 Method and equipment for eliminating echo
CN111756906A (en) * 2020-06-24 2020-10-09 上海立可芯半导体科技有限公司 Echo suppression method and device for voice signal
CN112201267A (en) * 2020-09-07 2021-01-08 北京达佳互联信息技术有限公司 Audio processing method and device, electronic equipment and storage medium
CN112545547A (en) * 2020-11-25 2021-03-26 北京积水潭医院 Breath sound analysis method and breath sound analysis system
CN112837697A (en) * 2021-02-20 2021-05-25 北京猿力未来科技有限公司 Echo suppression method and device
CN113194387A (en) * 2021-04-27 2021-07-30 北京小米移动软件有限公司 Audio signal processing method, audio signal processing device, electronic equipment and storage medium
CN113286047A (en) * 2021-04-22 2021-08-20 维沃移动通信(杭州)有限公司 Voice signal processing method and device and electronic equipment
CN113345469A (en) * 2021-05-24 2021-09-03 北京小米移动软件有限公司 Voice signal processing method and device, electronic equipment and storage medium
CN113489854A (en) * 2021-06-30 2021-10-08 北京小米移动软件有限公司 Sound processing method, sound processing device, electronic equipment and storage medium
CN114664322A (en) * 2022-05-23 2022-06-24 深圳市听多多科技有限公司 Single-microphone hearing-aid noise reduction method based on Bluetooth headset chip and Bluetooth headset

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101901601A (en) * 2010-05-17 2010-12-01 天津大学 Method and system for reducing noise of voice communication in vehicle
CN101917527A (en) * 2010-09-02 2010-12-15 杭州华三通信技术有限公司 Method and device of echo elimination
CN103327201A (en) * 2012-03-20 2013-09-25 联芯科技有限公司 Elimination method and system of remaining echoes
CN104427143A (en) * 2013-09-06 2015-03-18 联芯科技有限公司 Residual echo detection method and system
US20160063997A1 (en) * 2014-08-28 2016-03-03 Audience, Inc. Multi-Sourced Noise Suppression

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101901601A (en) * 2010-05-17 2010-12-01 天津大学 Method and system for reducing noise of voice communication in vehicle
CN101917527A (en) * 2010-09-02 2010-12-15 杭州华三通信技术有限公司 Method and device of echo elimination
CN103327201A (en) * 2012-03-20 2013-09-25 联芯科技有限公司 Elimination method and system of remaining echoes
CN104427143A (en) * 2013-09-06 2015-03-18 联芯科技有限公司 Residual echo detection method and system
US20160063997A1 (en) * 2014-08-28 2016-03-03 Audience, Inc. Multi-Sourced Noise Suppression

Cited By (30)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107393550A (en) * 2017-07-14 2017-11-24 深圳永顺智信息科技有限公司 Method of speech processing and device
CN109256145A (en) * 2017-07-14 2019-01-22 北京搜狗科技发展有限公司 Audio-frequency processing method, device, terminal and readable storage medium storing program for executing based on terminal
CN109256145B (en) * 2017-07-14 2021-11-02 北京搜狗科技发展有限公司 Terminal-based audio processing method and device, terminal and readable storage medium
CN107393550B (en) * 2017-07-14 2021-03-19 深圳永顺智信息科技有限公司 Voice processing method and device
CN107393546A (en) * 2017-09-04 2017-11-24 恒玄科技(上海)有限公司 A kind of echo cancel method and speech recognition apparatus for speech recognition process
CN109658946A (en) * 2017-10-12 2019-04-19 深圳前海黑鲸科技有限公司 A kind of echo processing method, device, storage medium and terminal device
CN108696648A (en) * 2018-05-16 2018-10-23 北京小鱼在家科技有限公司 A kind of method, apparatus, equipment and the storage medium of Short Time Speech signal processing
CN110176244A (en) * 2018-06-19 2019-08-27 腾讯科技(深圳)有限公司 Echo cancel method, device, storage medium and computer equipment
CN110176244B (en) * 2018-06-19 2023-10-03 腾讯科技(深圳)有限公司 Echo cancellation method, device, storage medium and computer equipment
CN111402910A (en) * 2018-12-17 2020-07-10 华为技术有限公司 Method and equipment for eliminating echo
CN111402910B (en) * 2018-12-17 2023-09-01 华为技术有限公司 Method and equipment for eliminating echo
CN109785853B (en) * 2019-03-11 2020-06-16 出门问问信息科技有限公司 Echo cancellation method, device, system and storage medium
CN109785853A (en) * 2019-03-11 2019-05-21 出门问问信息科技有限公司 A kind of echo cancel method, device, system and storage medium
CN110021289A (en) * 2019-03-28 2019-07-16 腾讯科技(深圳)有限公司 A kind of audio signal processing method, device and storage medium
CN110021289B (en) * 2019-03-28 2021-08-31 腾讯科技(深圳)有限公司 Sound signal processing method, device and storage medium
CN110418233A (en) * 2019-07-26 2019-11-05 歌尔股份有限公司 A kind of earphone noise-reduction method, device, earphone and readable storage medium storing program for executing
CN111048118A (en) * 2019-12-24 2020-04-21 大众问问(北京)信息科技有限公司 Voice signal processing method and device and terminal
CN111048118B (en) * 2019-12-24 2022-07-26 大众问问(北京)信息科技有限公司 Voice signal processing method and device and terminal
CN111756906A (en) * 2020-06-24 2020-10-09 上海立可芯半导体科技有限公司 Echo suppression method and device for voice signal
CN112201267A (en) * 2020-09-07 2021-01-08 北京达佳互联信息技术有限公司 Audio processing method and device, electronic equipment and storage medium
CN112545547A (en) * 2020-11-25 2021-03-26 北京积水潭医院 Breath sound analysis method and breath sound analysis system
CN112837697A (en) * 2021-02-20 2021-05-25 北京猿力未来科技有限公司 Echo suppression method and device
CN113286047A (en) * 2021-04-22 2021-08-20 维沃移动通信(杭州)有限公司 Voice signal processing method and device and electronic equipment
CN113286047B (en) * 2021-04-22 2023-02-21 维沃移动通信(杭州)有限公司 Voice signal processing method and device and electronic equipment
CN113194387A (en) * 2021-04-27 2021-07-30 北京小米移动软件有限公司 Audio signal processing method, audio signal processing device, electronic equipment and storage medium
CN113345469A (en) * 2021-05-24 2021-09-03 北京小米移动软件有限公司 Voice signal processing method and device, electronic equipment and storage medium
CN113489854A (en) * 2021-06-30 2021-10-08 北京小米移动软件有限公司 Sound processing method, sound processing device, electronic equipment and storage medium
CN113489854B (en) * 2021-06-30 2024-03-01 北京小米移动软件有限公司 Sound processing method, device, electronic equipment and storage medium
CN114664322A (en) * 2022-05-23 2022-06-24 深圳市听多多科技有限公司 Single-microphone hearing-aid noise reduction method based on Bluetooth headset chip and Bluetooth headset
CN114664322B (en) * 2022-05-23 2022-08-12 深圳市听多多科技有限公司 Single-microphone hearing-aid noise reduction method based on Bluetooth headset chip and Bluetooth headset

Also Published As

Publication number Publication date
CN106898359B (en) 2020-03-17

Similar Documents

Publication Publication Date Title
CN106898359A (en) Acoustic signal processing method, system, audio interactive device and computer equipment
KR101934636B1 (en) Method and apparatus for integrating and removing acoustic echo and background noise based on deepening neural network
CN109727604B (en) Frequency domain echo cancellation method for speech recognition front end and computer storage medium
EP1080465B1 (en) Signal noise reduction by spectral substraction using linear convolution and causal filtering
CN104158990A (en) Method for processing an audio signal and audio receiving circuit
KR101737824B1 (en) Method and Apparatus for removing a noise signal from input signal in a noisy environment
CN109979476B (en) Method and device for removing reverberation of voice
CN108172231A (en) A kind of dereverberation method and system based on Kalman filtering
CN108010536B (en) Echo cancellation method, device, system and storage medium
CN110211602B (en) Intelligent voice enhanced communication method and device
CN112037809A (en) Residual echo suppression method based on multi-feature flow structure deep neural network
Gil-Cacho et al. Transform domain prediction error method for improved acoustic echo and feedback cancellation
CN111223492A (en) Echo path delay estimation method and device
Yang et al. A noise reduction method based on LMS adaptive filter of audio signals
CN109727605A (en) Handle the method and system of voice signal
CN109559756A (en) Filter factor determines method, echo cancel method, related device and equipment
CN104410762A (en) Steady echo cancellation method in hand free cell phone conversation system
US20160019906A1 (en) Signal processor and method therefor
JP2003309493A (en) Method, device and program for reducing echo
JP3673727B2 (en) Reverberation elimination method, apparatus thereof, program thereof, and recording medium thereof
CN111370016B (en) Echo cancellation method and electronic equipment
Chhetri et al. Regression-based residual acoustic echo suppression
Tong et al. Acoustic echo suppression based on speech presence probability
CN111883155A (en) Echo cancellation method, device and storage medium
CN112687285B (en) Echo cancellation method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
PE01 Entry into force of the registration of the contract for pledge of patent right

Denomination of invention: Audio signal processing method, system, audio interaction device and computer equipment

Effective date of registration: 20230223

Granted publication date: 20200317

Pledgee: China Construction Bank Corporation Shanghai No.5 Sub-branch

Pledgor: SHANGHAI XIAOI ROBOT TECHNOLOGY Co.,Ltd.

Registration number: Y2023980033272

PE01 Entry into force of the registration of the contract for pledge of patent right