TW201110108A - Voice noise elimination method for microphone array - Google Patents

Abstract

A voice noise elimination method for microphone array applicable for voice enhancement and voice identification is disclosed. A microphone array and a noise elimination method are used to enhance signal-to-noise ratio (SNR) to achieve goals of effectively improving voice communication quality and voice identification rate. The method includes a voice and non-voice section detection technology and two stages of voice enhancement technology: (1) collecting voice signals in the space by a linear microphone array and processing the voice signals for determination of voice and non-voice sections; (2) effectively evaluating a long-term noise spectrum of the non-voice section to eliminate the noise on the voice signal spectrum, which is considered a first voice SNR enhancement; and (3) achieving the result of the second stage, which is also the final stage, of voice SNR enhancement by residue noise estimation and elimination complying with voice signal partial characteristics.

Description

201110108 VI. Description of the Invention: [Technical Field of the Invention] The present invention is directed to a voice hetero-touching method, which is to search for a voice signal in a microphone array to perform noise in a voice signal. The method can be applied to the field of voice signal processing technologies such as speech enhancement and speech recognition. [Prior Art] In recent years, the products advertised as smart mobile devices have brought a lot of convenience to users, and the voice input is even more or less in such product towels. The finer can make the material _ secret wire set communication number 'through (4) The ray channel and the loop wire make the listening through the axis. Fine, versatile transmission channels and generalized destruction of voice products, speech _ rate, shame development, and the obvious conversion technology for speech recognition is an important issue. In terms of technology, there is a noise elimination method in the frequency domain, and a 2-spectral subtraction method suppresses noise, and the miscellaneous_ is _sentence or _segment (four) average two 峨 _ _, "get it, the reason is to take care of the pepper There are still many shortcomings in the effect. There is still a lot of deficiencies in the way of things. It is not a good one. The inventor of this case is new in view of the shortcomings of the above-mentioned f-square handles, and after many years of hard work, rT1>A> Improve the innovation of noise and noise to eliminate green. _ After the research, finally successfully developed the county piece money wind array language [invention content] The object of the present invention is to provide a microphone array voice signal processing technology to achieve the age of H 曰 removal method, is to use the local voice unique New Zealand ___ rate under the circumstances. 1岐善杂讯环 201110108 二=:=::=:::= Improve voice communication quality and voice recognition recognition rate. ' ',处乂[Implementation] Main Z--shown as a schematic diagram of the microphone array voice noise elimination method of the present invention, a discourse;: a delay paste _ number - (four) measurement mechanism 3: an array voice processing mechanism 1 Generated time-lapse segment information TM ^ far and distance speaker's recent microphone speech signal 2 discriminates non-speech time miscellaneous mechanism 4: the speech signal after delaying it with time is processed by long == system 4 - Phase-enhanced speech signal, completes the first-stage noise cancellation with the long-term noise spectrum estimated by the non-speech segment for a long time; slave-local noise elimination mechanism 5: the first ρ-segment 5 The treatment was awarded to the first stage of the sputum plantation by the local noise to eliminate the Μ Gan + tiger, as a result of the final elimination of speech noise. No., and for (4) miscellaneous _ making two or more microphones simultaneously recording voice signals and can be linearly arranged or non-linearly arranged. Refer to FIG. 2, which is a schematic diagram of processing of the array voice L唬 processing mechanism of the microphone array voice noise cancellation method of the present invention, and the steps include: Step 1: 7 wind_sound signal 丨 骑 骑 骑 骑 估 估 u u u The time delay of the microphone m receiving the voice every microphone, wherein the time delay estimation is handled by Guan Sun (4), and all the microphones in the microphone _ _ The time delay of receiving the voice 201110108 Step 2: Re-voice money superimposition processing 12 will superimpose all the microphones when the late riding voice signal is superimposed, the combined voice signal, the voice recording superimposed processing, the system utilizes each The microphone receives the time delay of the voice, and superimposes the voice signals of all the microphones to obtain the combined voice signal; Step 3: the same as the sister _ voice minus the voice of the nearest speaker, 13 can be used to reach the distance speaker The voice signal of the microphone is recently selected to process the speech signal of the nearest speaker, and the recent speech of the speaker is processed by the money, and the time delay of receiving the voice from the side of the money Determine the distance of the nearest speaker-microphone speech signal; Step 4: Finally, the time delay of the speech signal and the microphone closest to the speaker's voice and after superimposed

The signal is respectively subjected to Fourier transform (FFT) to obtain the relative voice signal Y. Please refer to FIG. 3, which is the voice/non-intelligent section of the microphone array voice noise cancellation method of the present invention. According to the _ series and private voice dumping discs = the good spectrum of the near-speaker New Zealand's good spectrum money for each - (four) _ fresh voice 31 calculation, as shown in formula (1):

Ca/)=J2D£L£MVK〇(/)f-|z)s,(/)f Ο) where the mathematical symbols are as follows: M (0 〇 the voice signal τ recorded by the first microphone) The frequency spectrum signal recorded by the microphone is delayed by the delay time estimation process 1 time delay and the overlapped speech spectrum signal of the /th frame = correlation coefficient and threshold value of each tone Compared with the voice and the non-201110108 V, the long-term noise frequency 1# estimation process 42 will be used for the non-speech segment, and the 彳__ spectrum will be obtained; Then, it is combined with the speech signal (2) shown in Fig. 2. The formed speech spectrum signal is subjected to the audio spectrum signal spectrum elimination processing 41, that is, the first stage enhanced language (2) is obtained. ^(/)| =|D5;(/)| -p(/)l where the mathematical symbols are as follows. ^,(/) έ_ ^ ' ^(/) The sound of the 'sound box' Spectral signal sw) The long-term noise spectrum estimated in the two non-speech segments is shown in Figure 5. The first phase of the mechanism is to enhance the speech spectrum to eliminate the signal processing. Method The local noise signal, the noise spectrum of the non-speech segment before and after the sound frame is hidden by the sound/phase (for example, the appropriate local noise is estimated by the buffer spectrum estimation processing 52). The t-shirt r Filtering method is used to estimate it, and then it is stronger than the first-order: 2: the spectral signal is subjected to the noise suppression part of the local two-tone signal as shown in equation (3). Obtained 5 of them, mathematical symbols such as (3) SS, (f) one (/) to eliminate the mechanism after the first stage of enhanced speech spectrum signal 咚 (7) soil base 'local miscellaneous estimate by Liu Fei voice segment The second phase of the spectrum after the local Wei elimination mechanism enhances the speech spectrum signal. The second _enhanced speech spectrum of the Jit-passing Jit passes through ^,)' to completely eliminate the final speech signal after the noise. The microphone of the microphone ship is removed by the method ’’’. The implementation diagram of the daily 6 signal is arranged in a space of 4 square meters in a conference room environment, arranged with a linear microphone array 7, 4 ι〇 cm equidistant linear arrangement 201110108 microphone 7ι 'set talker distance microphone array 7 left front 3 cm and the microphone array 7% 3G degree scales sound. The speech signal of the speech of the present embodiment is subjected to the two I5 white & n of the present invention. The 曰 enhancement method can achieve effective noise cancellation. The microphone array voice noise cancellation method provided by the invention has the following advantages when compared with other technologies: 1. The invention proposes a voice detection method based on the gram wind, which can reach the effective voice of the single Cutting effect with non-speech segments.

2. The invention proposes a two-way voice strong financial method, which can achieve effective noise elimination effect. This month, a voice-based method based on a microphone array is proposed to improve voice quality. 4. According to the invention, (4) the voice recognition rate of the singularity of the wind gust and the voice side. The above description is directed to a specific embodiment of a possible embodiment of the present invention, but the embodiment is not intended to be used in the present invention, and is not equivalent to the equivalent implementation or modification of the present invention. , should be included in the patent scope t of this case. In summary, this case is not only innovative in terms of technical thinking, but also able to improve the quality of the attached items. It should be fully compliant with the progressive and progressive legal invention. The paste is poor according to the law. _, _^, To the sense of virtue. [Simplified illustration of the diagram] Figure: Schematic diagram of the method for eliminating the noise of the microphone array of the present invention; the processing diagram of the sound signal processing mechanism of the 膝 克 风 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Non-voice segmentation detector in addition to the long-term noise cancellation mechanism of the method except the local noise cancellation mechanism of the method r Figure 2 is a schematic diagram of the processing of the microphone array voice noise elimination according to the present invention; FIG. 5 is a schematic diagram of a microphone array voice noise cancellation 201110108 according to the present invention; and FIG. 6 is a schematic diagram of an implementation of a microphone array voice speaker voice signal of the microphone array voice noise cancellation method according to the present invention. [Main component symbol description] 1 array voice processing 2 microphone array voice signal 3 voice / non-voice segment detection mechanism 4 long-term noise cancellation mechanism 5 local noise cancellation mechanism 11 time delay estimation 12 voice superimposed 13 recent words Voice signal selection 31 voice correlation analysis 41 long-term noise spectrum cancellation 42 long-term noise spectrum estimation 51 local noise spectrum elimination 52 local noise spectrum estimation 6 speaker 7 microphone array 71 microphone

Claims (1)

201110108 VII. Patent application scope: 1. A method for eliminating voice noise in a microphone array, which mainly comprises: a distance-talker-most array voice processing mechanism, which generates a late-time and superimposed voice signal and a near-microphone voice signal; - The tone/non-speech segmentation mechanism' is a time delay in generating the array speech processing mechanism. The recent microphone signal of the sa domain and the translator's microphone voice segment-long-term noise cancellation mechanism completes the first-stage noise cancellation with the long-term noise spectrum estimated by the long-term non-speech segment; A local noise mechanism, the local noise spectrum estimated by the non-speech segment before and after the speech segment completes the second phase of noise cancellation. 2. The microphone array voice method according to the application of the fourth paragraph of her application, which allows the microphone array to make (four) or the community's mic period voice signal. 3. The method for eliminating the microphone array voice noise according to the application of the fourth patent, wherein the method of manufacturing the microphone array can be a linear lake or a non-linear arrangement. 4. For example, the microphone _ voice noise shame, which is described in detail, includes the steps of the array voice processing mechanism including: _ Step 1: Using the delay time estimation process to calculate the voice received by each microphone in the microphone array Time delay; Step 2: using voice signal superposition processing to superimpose different time delay speech signals of all microphones to obtain combined voice signals; Step 3: simultaneous microphone array voice signals are selected through recent speaker voice signal selection processing The speaker's recent microphone voice signal is used to perform the recent speaker voice signal selection process; Step 4: Finally, the time delay and the superimposed voice signal and the voice signal closest to the speaker microphone are respectively subjected to Fourier transform (FFT). Relative to the speech spectrum signal. 201110108 5. The microphone array voice noise cancellation method according to claim 4, wherein the step 1 delay time estimation process uses the relative position of each microphone to the speaker in the microphone array. Calculate the time delay in which the microphone is disconnected from each microphone. 6. The microphone array voice noise cancellation method according to claim 4, wherein the step 2 voice signal superimposition processing uses the time delay of each microphone to receive the voice, and superimposes the voice signals of all the microphones. The combined voice signal is processed. 7. In the case of the microphone array voice noise elimination green 所述 described in item 4 of the patent application, step 3, the recent speaker voice signal selection process is to use the time delay of each voice received by the microphone to determine the nearest speaker. Microphone voice signal. 8·If you apply for the general section of the i-Lin 〗 〖Speaking _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ The speech spectrum signal closest to the speaker's microphone is calculated, and the band correlation coefficient value of the two is calculated. Then the value is determined by the threshold value to determine that the speech series is a broken speech segment. Microphone_Vehicle Noise Age Method The steps of the noise cancellation mechanism include: Step 1: Using long-term noise spectrum estimation processing to estimate the time-frequency noise spectrum of multiple non-speech segments; Step 2: The long-term noise spectrum and the time-delayed and superimposed speech spectrum signal 1 are known to be numerated by the μ noise, and the _ _ speech spectrum signal can be obtained. Microphone array voice The signal elimination method, the complex elimination mechanism 'system _-stage increases (10) the speech spectrum signal, and then the peri-second operation, and the local miscellaneous/de-calculation estimated by the non-speech segment, and finally the second stage is obtained. Enhanced voice touch = residual miscellaneous money