TW526468B - System and method for eliminating background noise of voice signal - Google Patents

System and method for eliminating background noise of voice signal Download PDF

Info

Publication number
TW526468B
TW526468B TW90125856A TW90125856A TW526468B TW 526468 B TW526468 B TW 526468B TW 90125856 A TW90125856 A TW 90125856A TW 90125856 A TW90125856 A TW 90125856A TW 526468 B TW526468 B TW 526468B
Authority
TW
Taiwan
Prior art keywords
speech
background noise
filter
signal
voice signal
Prior art date
Application number
TW90125856A
Other languages
Chinese (zh)
Inventor
Jia-Hung Liou
Original Assignee
Chunghwa Telecom 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 Chunghwa Telecom Co Ltd filed Critical Chunghwa Telecom Co Ltd
Priority to TW90125856A priority Critical patent/TW526468B/en
Application granted granted Critical
Publication of TW526468B publication Critical patent/TW526468B/en

Links

Abstract

A system and method for eliminating background noise of voice signal is provided, which combines the adaptive filter with long-term and short-term statistical characteristics for voice. Because the statistical characteristics of voice signal are varied with the time, the associated coefficients of the filter also need to be suitably adjusted along with the voice signal, so as to eliminate the undesirable background noise, and further compensate the high frequency deterioration by the filter to increase the brightness of the sound and obtain the best voice signal.

Description

A7

V. Description of the Invention (Technical Field) The present invention relates to a system and method for suppressing voice signal f scene noise suppression > 'It is mainly a background noise suppression system and method designed for speech short time and time correction. [Previous technology] Voice signals are the most important data type capability transmitted in communication systems. In addition to voice signals during communication, background noise in the call environment will also accompany the phone, which will cause a certain degree. Interference, which in turn affects call quality; especially recently the rapid growth of mobile communication phones, f is tolerated by background noise, so the technology to suppress background noise is an important issue in communication systems that emphasize service quality There are three commonly used background noise suppression techniques: The first method is the noise cancellation method in the frequency domain. The basic method of this method is in the non-speech section. It is estimated that the noise in the frequency domain at this time = 2 The level in the next speech segment is subtracted from the solution field: the estimated noise energy at each frequency. This method is simple, but- The statistical characteristics of general background noise will change over time, so its effect of suppressing background noise is limited. In US Patent No. 2 and US05 = 2927, the concept of using the frequency domain noise cancellation method is mentioned. The two methods are the time-domain noise cancellation method. The principle of this method is to use two microphones to receive external signals. The first microphone ^ is mainly used to receive signals of voice and background noise, and the first one is the signal of scene noise. Therefore, with the second microphone, the size of the monthly news is displayed, and then the U ruler of the first microphone is used in the time field. Cai 10 15 20 (Please read the precautions on the back before filling this page) «- -------- Line One ____ 一 ___.__, —I—I—I—IIIII ✓ IIII. 526468 A7

526468 A7 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. The quality of the invention description (&gt;), and the correlation coefficient can be adjusted adaptively. Another object of the present invention is to provide a system and method for suppressing background noise of a voice signal, which has low complexity and requires only a microphone, so it is quite suitable for application in recent fast-growing mobile phones and speech recognition technology. To improve speech encoding quality and speech recognition rate. [Technical content] The background noise suppression system and method for speech signals are used to improve the degradation of speech quality caused by the background noise. The analog speech signals are first converted by the sampler to analog digital signals for subsequent use. For digital signal processing. The bandwidth of the voice signal is about 4KHz. According to the Nyquist sampling theorem, the general sampling frequency is 8KHz. In order to improve the correlation between the sampling signals, we increase the sampling frequency by four times, that is, the sampling frequency of 32KHz, which is called super sampling. The digital signal after sampling is expressed by 12-bit Pulse Code 15 Modulation (PCM). That is to say, the allowable change range of digital speech sampling is within ± 2048. The background noise suppression system and method for speech signals of the present invention include: an oversampling unit, two low-pass wave waver units, an adaptive speech analyzer unit, a speech period detector unit, and a background noise 20 signal Suppression filter unit, and a high-frequency booster unit. Assume that the voice signal containing background noise is a cross (0; first 乂 (0 will be oversampled by the oversampling unit (more than twice the sampling frequency of speech and audio width), and the sampled digital signal will be &amp; ⑷, expressed by 12-bit pulse wave code modulation, where A represents the first sampling signal. Because of the oversampling ______- 5- _ This paper size is applicable to China National Standard (CNS) A4 specification (21G X 297)) (Please read the precautions on the back before filling out this page) -Installation · —Line — 526468 A7 V. Description of Invention (仏) Printing Department of Employees' Cooperatives, Intellectual Property Bureau, Ministry of Economic Affairs, outside of voice signal bandwidth Noise will also be included; therefore, after the oversampling unit, a low-pass filter must be passed to remove unnecessary signals outside the bandwidth of the voice signal. The digital signal after the first low-pass filter, I⑷ , Into the adaptive speech analyzer unit, speech cycle detector unit 70, and background noise suppression filter unit for the next step of processing. In the adaptive speech analyzer unit, An all-pole adaptive wave filter of order &quot; is used to estimate the speech signal, and the coefficient of the whole device is _, a} represents the first waverbinator coefficient = === the filter coefficients representing the unique characteristics of the voice signal, will Π) reaches the month, 'the noise suppresses the waver unit; on the other hand, it will be sent to the speech period detector unit to estimate the period of the speech signal, and the estimated period P ranges from 3 to 10w, if The sampling frequency is a duty, then the number of samples corresponding to a cycle is about 96 ~ 32. The sampling cycle of each voice signal will be estimated, and it will be sent to the background with the 15 coefficients of the all-pole adaptive wavelet. The noise suppression wave filter unit is used to suppress the background noise in the next step. Li Ba ί !! The noise suppression wave filter unit 'uses the adaptive filter: and the speech and period detector units to estimate the obtained filter coefficients, respectively. The k / ρ 5 is also considered as the background noise suppression filter. After passing through the first 20 low-pass data wave filters, the background noise 2 designed by us is now sent to reduce the background doped in the speech signal. Noise energy is not compared. The high-frequency component in the original voice signal will also be suppressed by the two = sounder, so we have designed a high-frequency __ 吾 音 signal high-frequency suppressed component. Finally, according to this paper scale, the Chinese national standard is applied ? 5 line news 526468 A7 10 15 Printed invention description printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs (^) Pass a low-pass filter to filter out noise outside the bandwidth of the voice signal and get a voice signal with improved quality. [Brief description of the drawings]. Please read the following detailed description of a preferred embodiment of the present invention and its accompanying drawings ^ You can further understand the technical content of the present invention and its purpose and effectiveness; the drawings related to this embodiment For: Figure-This is the architecture diagram of the background noise suppression system of the voice signal of the present invention; Figure 2 is a circuit block diagram of the background noise suppression system analyzer of the voice signal; 'The adaptive prediction of the system is a block diagram of the coefficient estimation circuit; Figure 4 is a circuit block diagram of the device for the background noise suppression system of the voice signal; and ^ Double test chart 5 is the voice signal King noise suppression, Kun, Shi Α ~ ,,, view of the first noise suppression Gaoshu Shu filtering circuit block diagram. [Symbols of main parts] 101 Supersampler 102 Low-pass filter 103 Adaptive speech analyzer 104 Background noise suppression filter 105 Speech period detector 106 South frequency enhancer 107 Low-pass filter 21 Positive and negative discriminator (please (Please read the precautions on the back before filling out this page) Loading ir ° J. This paper ruler? Guan Jiaxian (CNS) A4 specification 297 mm) 526468 Α7 __ Β7 V. Description of the invention (PA0IG2fle.TWP-8Π 0 10 Printed by the Intellectual Property Bureau Employee Consumer Cooperative of the Ministry of Economic Affairs 15 22 Step estimator 221 Adaptive step decision Filter 23 adaptive estimation filter 31 positive and negative discriminator 41 period detector 51 noise shaping filter [preferred embodiment] Please refer to FIG. 1, the voice signal background noise suppression system of the present invention includes: Supersampler 101, two low-pass filters 102, 107, an adaptive speech analyzer 103, a speech period detector 105, a background noise suppression filter 104, and a High-frequency intensifier 106. Before the background noise suppression process is officially carried out, it will be pre-processed to convert the analog voice signal into a digital signal suitable for subsequent processing, including oversampling and low-pass filtering, wave, and oversampling. On the one hand, the converter 101 performs analog digital conversion on the external analog voice signal. On the other hand, the converted digital signal is expressed by Pulse Code Modulation (JPCM) method. When changing, the sampling frequency is much larger than the minimum frequency specified by the sampling theorem, in order to improve the correlation between the samples. In this embodiment, the sampling frequency is recommended to be 32 times fe, which is 8 times the general voice and audio frequency. The low-pass filter 102 is used to remove noise outside the speech and audio width, especially after passing through the supersampler 101, it is necessary to use a low-pass filter 102 to limit the bandwidth of the signal. Within the bandwidth to improve the performance of the subsequent processing unit. In the present embodiment, 'a third-order Butterworth low-pass filter is used, and the cut-off frequency is set to 4 voice bands of speech. The signal after low-pass filtering (Please read the notes on the back before filling in this page) «. Order: i-line-(cns) a4 m ^ 10 x 297 ^ t) — --- ~ 526468 Printed by the Consumers’ Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 、 Explanation of invention (2) &quot; PAO ICUA ^ '. TWP-y / la l (A〇 knife enters adaptive speech analyzer 103, speech period detector 105, moon hall, noise suppression filter 1 〇6, for the next stage of processing-Adaptive speech analyzer circuit Block diagram. The adaptive speech analysis 103 includes a positive-negative discriminator 21, a step estimator 22, and an adaptive estimation filter 23. The positive-negative discrimination H21 is based on the input speech signal I㈨ and the adaptive speech analyzer 103. The estimated size of & ㈨ determines the value of output bit 6 (A). B (k) ^ fSnn (k) &gt; Se (k) ^^ nn (^) &lt; Se (k) 步 step The order estimation of 22 is to use the previously determined bits to estimate the current sampling step. This step represents the compensation of the estimated residual of the adaptive prediction filter 23. Assuming the currently determined bits are commanded), the adaptive step determiner in step estimator 22 will be based on commanded) and its first three bits commanded -1), 蚴 -2), 蚴-Xi To determine the current state of the adaptive speech analyzer, and determine a modified coefficient, as shown in Table 1. After that, the feedback averager of one order is used to generate the step imagination estimated at time ^ as follows: 5 {k)-5 (k + a (k) (2), where /? &Lt; 1 is the constant of the feedback average line, used to control the average length. The closer P is to 1, the longer the average length, but the slower the reaction time, the general size is about 0.9. Fly is used to adjust the correction factor. ), So that the adaptive speech analyzer 103 can follow the change of the speech signal. Finally, the γ-order adaptive estimation filter 23 uses the past &quot; Phrase 1 This paper size is applicable _ ^ Standard (CNS) A4 specifications (21Q χ 297 public [10 15 20 (Please read the precautions on the back before filling (This page) | Line · 526468 A7 V. Description of the invention (nQ2QQ--TWP-10/10 10 15 * The 20-tone estimate and the step estimate MU are printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs to generate the next voice The estimated value of the sample is hi) ^ N- \ ^ k ^) ^ ai (krsu (ki) + s (k) (3) Table 1 is a reference table of the adaptive step determiner 221. The modified coefficient α (々 ) Is determined according to this table. If the four consecutive bits are all the same, it means that & 估计 estimated by the adaptive speech analyzer 10 is not large enough, so the correction coefficient &lt; magic is set to 2 , So that the adaptive speech analyzer 103 can quickly catch up with changes in the speech signal. If only three consecutive bits are the same, give a slightly smaller correction coefficient α (magic = 1, to slightly increase the step Size. If four consecutive bits are not the same, set the correction factor to _ 丨. Because at this time The adaptive speech analyzer 103 estimates the speech signal, so it must be corrected. In the rest of the cases, <Magic = 0 'represents the change in the speech signal tracked by the adaptive speech analyzer 103. Figure 3 shows the adaptive prediction. Filter 23 coefficient estimation circuit block diagram for generating # coefficients of the W-order adaptive estimation filter 23,% ,, / = 1, 2, ...., #. Adaptive estimation filter 23 coefficient estimation Contains a positive and negative discriminator 31, two rows of #_ 丨 tapped delay lines, and a first-order feedback averager of length iV. Two input signals, including the estimated value of the voice signal 5; ㈨ and bits The estimated value of the voice signal: before entering the upper row of tapped delay lines, it will first pass the positive and negative discriminator 31, and will be given +1, one according to its positive and negative sign; []. This can reduce the amount of calculation. Bit value 0 (phantom will be multiplied by a μ's ruler before entering the next row of tapped delay lines? (CNS) A4ii ^ -10- χ 297 ^ 1) (Read the note on the back first (Please fill in this page for matters) __ ----- ^ ------------------------------------ 526468

Description of the invention (1 is less than 1 constant gain e. V output signals, that is, ~ coefficients of the adaptive estimation filter 23, orbit / = 1, 2, ... ^. Take the / coefficient _ as an example ( z # 1), we can use the following equation to show how it is generated: 5 = ^ ~ 1) + e * b (k) * SGN [Se (k)] (4) where d is a constant, representing The average length of the first-order feedback averager is generally about 0.9. SGN [] represents the action of the positive and negative discriminator, that is, the positive and negative signs sampled in square brackets. Basically, equation (4) is a simplified search method based on the stochastic gradient minimization. In the method of generating the λM solid coefficient of the adaptive pre-estimation filter 23, the method of generating% ㈨ has a constant term γ less than 1: γ: a, (k) = -1) + e * b (k) * SGN [Se (k)] + / (5. This is to indicate that there is a strong correlation between the currently estimated speech signal and the previous estimate of the speech signal. 15 Figure 4 is a block diagram of the speech period detector circuit, used to estimate The period of the speech signal. The speech period detector 105 includes a row of tapped delay lines of length (κ, η + ι), (Kin + 1) subtracters, (Unin +1) absolute value units, +1) a first-order feedback averager, and a period detector 41. ^ Represents the maximum possible period of speech 20 periods, 4 represents the minimum possible period of speech. If the sampling frequency is 32 NZ, then 4,32, so the number of lengths of the tapped delay line, subtractor, absolute value unit, and first-order feedback averager is 225. The input signal of the speech period detector 105 ㈣ stores the past (n + 1) values through the tapped delay line on the one hand, and-on the other hand ㈣-11 paper rule money and wealth ~~ --- (please (Please read the notes on the back before filling this page) · i-line-Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs, Consumer Cooperatives 526468 A7

After the value of tjt is subtracted and the absolute value is taken, they respectively enter the -step back, == The above action is to find the correlation between the input and past speech estimates ▲〗 々. Assuming that 1⑷ has the most correlation with the first estimated value in the past, the output value of the first-order feedback averager corresponding to the P-th delay will be at most 5, so in the period detector 41, it will be based on equation (6) (7), detecting the desired speech period p, as follows: &quot;, tf —4 ^ m, /) |]) &gt; Danger, where E [] represents the average of a P white dagger feedback "{{(()}" 10 means to find the parameter that minimizes the value in parentheses. Fei is the critical value of the first-order feedback averager output value, an experience to distinguish vowels from non-vowels. value. If the current sample does not belong to the vowel in the voice signal, what is detected? = 〇15 Printed by the Intellectual Property Bureau's Consumer Cooperatives of the Ministry of Economic Affairs 20 (¾Read the precautions on the back before filling out this page) Figure 5 is a block diagram of the background noise suppression filter circuit, which is used to combine speech analysis and speech. Detect the speech characteristic coefficient you got) and the period of the speech, to perform background noise suppression. The background noise suppression filter H1G4 contains two rows of tapped delay lines with a length of #, one, and the delay is a chirp delay. , ... adders, _noise shaping wave is 5 input__ for voice signal, voice characteristics training), and voice period P, the output signal is the background noise signal slogan, which is Snn (kl ), Snn (k-2), 's ^ kN); The tapped delay line in the next row is based on the period p detected by the speech period detector. First, -_ 12 _ This paper standard applies Chinese national standards ( CNS) A4 regulation 297 publicly loved 526468 Α7 Β7 PAO 1028δ.Τν \ / Μ ^ 5 / Ιβ 5. Description of the invention (丨 f) The input signal is the voice signal, and the delay is p samples, and then the distance from the heart (Magic / &gt; Before the sampling and V voice signals before it, that is (please Read the notes on the reverse side and fill in this page) Λ〇. After that, ㈣ 人 ㈣, ......... and HPXSJk-P-V), ... A.-P-TV) this Add the two groups of signals in order, and then send them to the noise shaping filter 51 together with the speech characteristic coefficient 4. Since the similarity of the speech signals in these two groups of roars is quite high, it is Homological addition; on the other hand, background noise does not have this similarity, so it is non-homogeneous addition. Therefore, the effect of homophonic noise removal can be achieved (if p = 〇, this feature has no effect). The 10 waves of noise shaping filter are 51, and the iV +1 coherently added speech signal is combined according to the following transformation function of the shaping filter: 1-H (z) = — 弓-1-^ α, α / ζ '/ / = 1 • Line · where' α and 常数 two constants, 0 ^ Μ α ^ i, are used to control the size of the peaks and troughs of the speech spectrum. The closer α is to 丨 and 々 is closer to 0, then The peak printed by the Consumer Property Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs of the 15th, the larger the trough, the deeper, but the high-frequency signal will be more attenuated. Suggested choice疋 α = 0.9, 々 = 0 · 2. Because α i represents the characteristics of the voice signal, after the conversion function of the shaping filter 51, the spectrum of the original signal will be shaped into a shape similar to the voice signal, that is, the background The spectrum of noise will change with the spectrum of the voice signal, so the so-called masking effect 20 (MaskingEffect) is generated, and the effect of suppressing background noise is achieved. Since we have done the same operation before, we have significantly increased Improving the effect of shadowing effect —-----—-13 _ This paper ruler ® mesh home fresh (CNS) A4 test (21 ^ 7 ^-— ___ 526468 A7 V. Description of the invention ((?) Jzo0.TV \ ih -14 / Ιΰ 印 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs, the next step is to send the voice signal processed by the background noise suppression filter 104 to the high-frequency enhancer 106. ^ / (ζ) = \ -γζ ~ χ (9) Basically, this is a high-pass filter, 〇 &lt;? / &lt; 1, which is used to compensate for the mouth shape; consider the wave to reduce the frequency of agricultural noise reduction Impact. Finally, the same low-pass filter 107 as before is passed to remove noise other than the sound and audio width generated in the adaptive background noise suppression system. [Features and effects] The system and method for suppressing background noise of speech signals provided by the present invention have the following advantages when compared with the cited citations and other conventional technologies: ^, L The present invention provides a background noise background of speech signals The signal suppression system and method, on the one hand, use an all-pole linear prediction filter to reconstruct the speech signal, and the i $ 彳 plane also detects the period that only exists in the speech signal. After that, the background noise is suppressed based on the estimated correlation coefficient of the speech signal and the period of the speech, thereby improving the quality of the speech signal. 2 • The present invention provides a background noise suppression system and method for voice signals, which can greatly improve the quality of input signals with low signal-to-noise ratios, and can adaptively adjust correlation coefficients. 3. The present invention provides a background noise method for voice signals, which has low complexity and requires only I-φ ^ n H to be supported first, so it is quite suitable for μ-growth mobile phones and speech recognition technology. In the middle, borrow _ 张 尺 ^ t _ 家 鲜 10 15 20 (Please read the notes on the back before filling this page) -line- Γ I —r. 526468 A7 ---------------- -B7 _ Jade, invention description ((,) &quot; &quot; State_0 • The following mentions Nanan's speech encoding quality and speech recognition rate. Ming t ::: Detailed description is specific to the -feasible embodiment of the present invention 'However, the examples are not intended to limit the patent scope of the present invention. Any equivalent implementation or change that does not depart from the technical spirit of the present invention should be included in the patent scope of this 5 case. In summary, this case is not only in Technical thinking is an innovation, and can improve the above-mentioned multiple effects compared with conventional items. It should have fully met the statutory invention patent requirements of novelty and progress, apply according to law, and ask your office to approve this invention patent application to encourage invention. , Please. (Please read the notes on the back before filling (This page) Order: 丨 Line-Moderate Rule Paper _J this printed by the Consumers' Cooperative of Intellectual Property Bureau of the Ministry of Economic Affairs

526468 M 'printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 ----- PA010388.TWP 16/18 V. Description of the invention ((f) Table 1: Reference table for adaptive step determiner b (n) b (nl ) b (n-2) b (n-3) a (n) -1 -1 -1 -1 2 1 1 1 1 2 -1 -1 -1 1 1 1 1 1 -1 1 -1 1 1 1 1 1 -1 -1 -1 1 1 1 -1 -1 0 -1 -1 1 1 0 -1 1 1 -1 0 1 -1 -1 1 0 -1 -1 1 1 0 1 1 -1 _1 0 1 -1 1 1 0 -1 1 -1 -1 0 -1 1 _1 1 -1 1 -1 1 -1 -16- (Please read the precautions on the back and then fill out this page) Order ---- -Line 74: This paper size applies to China National Standard (CNS) A4 (210 X 297 public love)

Claims (1)

  1. 526468 Patent Application Fan Park_PAO 1Q288 TWP. 17/18 10 15 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 20 2. A method for suppressing background noise of voice signals, in which the analog voice flood is first passed through a sampler as an analog digital signal Conversion between samples; Sampling is performed at a sampling frequency of 32KHz, and the sampled digital flood number is represented by a 12-bit pulse wave code modulation. After sampling, it passes a low-pass filter to remove the bandwidth of the voice signal. Unnecessary signals' · The digital signal passing through the first low-pass filter enters the adaptive speech analyzer unit, the speech period detector unit, and the background noise suppression filter unit for further processing; In the adaptive speech analyzer, the full-pole adaptive transition of order # is used to receive the incoming voice signal, and the coefficient of the all-pole-fitting leather device is meaning (magic '/ = {1, n} represents the first solid Filter coefficients, these y filter coefficients' determined to represent the unique characteristics of the voice signal will be sent to the background noise suppression filter unit; another The face is sent to the speech period detector unit to estimate the period of the speech signal; the period of each speech signal sample is estimated and sent to the background noise suppression filter unit to suppress the background noise in the next step; in the background In the noise suppression filter, the background noise suppression filter is designed by using the filter number and the speech period, and then a high-frequency booster is used to compensate the high-frequency suppressed components of the speech signal. Finally, a low-pass filter is passed. Filter out noise outside the bandwidth of the voice signal.-A kind of background noise suppression of voice signals ", by the time t (CNS) A4 ^ m (210X2 ^ i ^ 526468
    Applicable patent scope ίο 15 and short-day interval Hf · characteristics, make appropriate adjustments based on changes in voice signals to eliminate unnecessary background noise; these include: Oversampling, which converts analog voice signals into digital voice Signals;-the low-pass filter, which removes unnecessary parts of the digital voice signal output by the supersampler;-adaptive gastric analysis II 'analyzes the digital voice output by the-low-pass filter Characteristics of the signal;-Speech period detectionϋ, which estimates the period of the digital speech signal output by the first-low-pass ferrule; a background noise suppression filter, which is based on the speech characteristics and speech period analyzed by the adaptive speech analyzer The speech period estimated by the detector is added to remove background noise. The high-frequency enhancer compensates for the attenuation of the digital speech signal caused by the background noise suppression filter. The low-pass filter is to remove the unnecessary part of the output of the high-frequency booster; by the above-mentioned components, the voice signal is processed to eliminate the In addition to the unnecessary background noise, the audio signal is compensated for the frequency reduction caused by the filter, so as to improve the brightness of the South sound to obtain the best voice signal. Mandatory ^ Applicable to China National Standard (CNS) Α4_ (210X297 mm (Please read the precautions on the back before filling this page)
TW90125856A 2001-10-19 2001-10-19 System and method for eliminating background noise of voice signal TW526468B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
TW90125856A TW526468B (en) 2001-10-19 2001-10-19 System and method for eliminating background noise of voice signal

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
TW90125856A TW526468B (en) 2001-10-19 2001-10-19 System and method for eliminating background noise of voice signal

Publications (1)

Publication Number Publication Date
TW526468B true TW526468B (en) 2003-04-01

Family

ID=28450658

Family Applications (1)

Application Number Title Priority Date Filing Date
TW90125856A TW526468B (en) 2001-10-19 2001-10-19 System and method for eliminating background noise of voice signal

Country Status (1)

Country Link
TW (1) TW526468B (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8069040B2 (en) 2005-04-01 2011-11-29 Qualcomm Incorporated Systems, methods, and apparatus for quantization of spectral envelope representation
TWI455119B (en) * 2006-07-24 2014-10-01 Marvell World Trade Ltd Rotating data storage device having audio monitoring module
US8892448B2 (en) 2005-04-22 2014-11-18 Qualcomm Incorporated Systems, methods, and apparatus for gain factor smoothing
TWI488179B (en) * 2008-06-30 2015-06-11 Audience Inc System and method for providing noise suppression utilizing null processing noise subtraction
US9536540B2 (en) 2013-07-19 2017-01-03 Knowles Electronics, Llc Speech signal separation and synthesis based on auditory scene analysis and speech modeling
US9558755B1 (en) 2010-05-20 2017-01-31 Knowles Electronics, Llc Noise suppression assisted automatic speech recognition
US9640194B1 (en) 2012-10-04 2017-05-02 Knowles Electronics, Llc Noise suppression for speech processing based on machine-learning mask estimation
US9799330B2 (en) 2014-08-28 2017-10-24 Knowles Electronics, Llc Multi-sourced noise suppression
US9830899B1 (en) 2006-05-25 2017-11-28 Knowles Electronics, Llc Adaptive noise cancellation

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8069040B2 (en) 2005-04-01 2011-11-29 Qualcomm Incorporated Systems, methods, and apparatus for quantization of spectral envelope representation
US8078474B2 (en) 2005-04-01 2011-12-13 Qualcomm Incorporated Systems, methods, and apparatus for highband time warping
US8140324B2 (en) 2005-04-01 2012-03-20 Qualcomm Incorporated Systems, methods, and apparatus for gain coding
US8244526B2 (en) 2005-04-01 2012-08-14 Qualcomm Incorporated Systems, methods, and apparatus for highband burst suppression
US8260611B2 (en) 2005-04-01 2012-09-04 Qualcomm Incorporated Systems, methods, and apparatus for highband excitation generation
US8332228B2 (en) 2005-04-01 2012-12-11 Qualcomm Incorporated Systems, methods, and apparatus for anti-sparseness filtering
US8364494B2 (en) 2005-04-01 2013-01-29 Qualcomm Incorporated Systems, methods, and apparatus for split-band filtering and encoding of a wideband signal
US8484036B2 (en) 2005-04-01 2013-07-09 Qualcomm Incorporated Systems, methods, and apparatus for wideband speech coding
US9043214B2 (en) 2005-04-22 2015-05-26 Qualcomm Incorporated Systems, methods, and apparatus for gain factor attenuation
US8892448B2 (en) 2005-04-22 2014-11-18 Qualcomm Incorporated Systems, methods, and apparatus for gain factor smoothing
US9185487B2 (en) 2006-01-30 2015-11-10 Audience, Inc. System and method for providing noise suppression utilizing null processing noise subtraction
US9830899B1 (en) 2006-05-25 2017-11-28 Knowles Electronics, Llc Adaptive noise cancellation
TWI455119B (en) * 2006-07-24 2014-10-01 Marvell World Trade Ltd Rotating data storage device having audio monitoring module
TWI488179B (en) * 2008-06-30 2015-06-11 Audience Inc System and method for providing noise suppression utilizing null processing noise subtraction
US9558755B1 (en) 2010-05-20 2017-01-31 Knowles Electronics, Llc Noise suppression assisted automatic speech recognition
US9640194B1 (en) 2012-10-04 2017-05-02 Knowles Electronics, Llc Noise suppression for speech processing based on machine-learning mask estimation
US9536540B2 (en) 2013-07-19 2017-01-03 Knowles Electronics, Llc Speech signal separation and synthesis based on auditory scene analysis and speech modeling
US9799330B2 (en) 2014-08-28 2017-10-24 Knowles Electronics, Llc Multi-sourced noise suppression

Similar Documents

Publication Publication Date Title
US8219389B2 (en) System for improving speech intelligibility through high frequency compression
US6263307B1 (en) Adaptive weiner filtering using line spectral frequencies
US7680652B2 (en) Periodic signal enhancement system
EP0790599B1 (en) A noise suppressor and method for suppressing background noise in noisy speech, and a mobile station
US8954324B2 (en) Multiple microphone voice activity detector
CA2571417C (en) Advanced periodic signal enhancement
EP1169883B1 (en) System and method for dual microphone signal noise reduction using spectral subtraction
JP4376489B2 (en) Frequency domain post-filtering method, apparatus and recording medium for improving the quality of coded speech
EP2056296B1 (en) Dynamic noise reduction
KR101052445B1 (en) Method and apparatus for suppressing noise, and computer program
US7941315B2 (en) Noise reducer, noise reducing method, and recording medium
JP4916394B2 (en) Echo suppression device, echo suppression method, and computer program
JP5302968B2 (en) Speech improvement with speech clarification
JP5068653B2 (en) Method for processing a noisy speech signal and apparatus for performing the method
US6070137A (en) Integrated frequency-domain voice coding using an adaptive spectral enhancement filter
CN101719969B (en) Method and system for judging double-end conversation and method and system for eliminating echo
US5706395A (en) Adaptive weiner filtering using a dynamic suppression factor
US8571231B2 (en) Suppressing noise in an audio signal
US7313518B2 (en) Noise reduction method and device using two pass filtering
RU2517315C2 (en) Method and device for audio signal processing
RU2127454C1 (en) Method for noise suppression
US9438992B2 (en) Multi-microphone robust noise suppression
US8315380B2 (en) Echo suppression method and apparatus thereof
EP1253581B1 (en) Method and system for speech enhancement in a noisy environment
ES2294506T3 (en) Noise reduction for automatic recognition of speech.

Legal Events

Date Code Title Description
GD4A Issue of patent certificate for granted invention patent
MM4A Annulment or lapse of patent due to non-payment of fees