EP1081685A2 - Rauschverminderungsverfahren in einem Sprachsignal mit einem einzigen Mikrophon - Google Patents

Rauschverminderungsverfahren in einem Sprachsignal mit einem einzigen Mikrophon Download PDF

Info

Publication number
EP1081685A2
EP1081685A2 EP00118147A EP00118147A EP1081685A2 EP 1081685 A2 EP1081685 A2 EP 1081685A2 EP 00118147 A EP00118147 A EP 00118147A EP 00118147 A EP00118147 A EP 00118147A EP 1081685 A2 EP1081685 A2 EP 1081685A2
Authority
EP
European Patent Office
Prior art keywords
noise
speech
data
blocks
block
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.)
Withdrawn
Application number
EP00118147A
Other languages
English (en)
French (fr)
Other versions
EP1081685A3 (de
Inventor
Russell H. Lambert
Karina L. Edmonds
Shi-Ping Hsu
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.)
Northrop Grumman Corp
Original Assignee
TRW Inc
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 TRW Inc filed Critical TRW Inc
Publication of EP1081685A2 publication Critical patent/EP1081685A2/de
Publication of EP1081685A3 publication Critical patent/EP1081685A3/de
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
    • G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
    • G10L21/0208—Noise filtering

Definitions

  • This invention relates generally to techniques for reliable conversion of speech data from acoustic signals to electrical signals in an acoustically noisy and reverberant environment.
  • ASR automatic speech recognition
  • background noise from both inside and outside an automobile renders in-vehicle communication both difficult and stressful.
  • Reverberation within the automobile combines with high noise levels to greatly degrade the speech signal received by a microphone in the automobile.
  • the microphone receives not only the original speech signal but also distorted and delayed duplicates of the speech signal, generated by multiple echoes from walls, windows and objects in the automobile interior. These duplicate signals in general arrive at the microphone over different paths.
  • multipath is often applied to the environment.
  • the quality of the speech signal is extremely degraded in such an environment, and the accuracy of any associated ASR systems is also degraded, perhaps to the point where they no longer operate.
  • recognition accuracy of an ASR system as high as 96% in a quiet environment could drop to well below 50% in a moving automobile.
  • speech compression Another related technology affected by noise and reverberation is speech compression, which digitally encodes speech signals to achieve reductions in communication bandwidth and for other reasons. In the presence of noise, speech compression becomes increasingly difficult and unreliable.
  • the active noise reduction approaches cancel acoustic noise signals by generating an opposite signal, sometimes referred to as "anti-noise,” through one or more transducers near the noise source, to cancel the unwanted noise signal.
  • This technique often creates noise at some other location in the vicinity of the speaker, and is not a practical solution for canceling multiple unknown noise sources, especially in the presence of multipath effects.
  • the present invention resides in a system and method for reducing noise in speech signals obtained from a single microphone in a noisy environment.
  • the present invention is a general noise reduction framework that allows multiple parameters to be adjusted optimally for any given application, noise environment or automatic speech recognition (ASR) system.
  • ASR automatic speech recognition
  • the system of the invention comprises a fast Fourier transform (FFT) circuit for transforming blocks of input microphone data to a frequency domain representation; a bandpass filter to remove selected frequency bands in which noise is known to be present; a speech detector for sensing the presence of speech signals in microphone data; a noise spectrum estimator updated only for data blocks in which no speech signals are detected; a spectrum subtraction circuit, for subtracting the estimated noise spectrum from microphone signals containing noise and speech signal components; and a speech emphasis circuit, for emphasizing speech signal components with respect to any residual noise after operation of the spectrum subtraction circuit, to provide a noise-reduced speech signal in the frequency domain.
  • FFT fast Fourier transform
  • the system may further comprise means for reconstructing time-domain data from the noise-reduced speech signal in the frequency domain, including an inverse fast Fourier transform circuit for transforming blocks of data from the frequency domain back into the time domain, whereby the noise-reduced speech signals are more intelligible in voice communication systems.
  • the system may further comprise an automatic speech recognition (ASR) system connected to receive the noise-reduced speech signals in the frequency domain, whereby the ASR system operates more reliably to generate selected control signals.
  • ASR automatic speech recognition
  • the speech emphasis circuit raises signals in the frequency domain by a power N, where N is a positive quantity greater than one.
  • the input signals are presented to the noise reduction system in blocks of "A” samples each, and data blocks of size “2A” samples each are presented to the FFT circuit.
  • the system further comprises means for combining input signal blocks of "A” samples in pairs to form data blocks.
  • the means for combining input signal blocks uses each input signal block twice, such that a currently input signal block is placed in a second half of a current data block and is then placed in a first half of a next data block.
  • the system may further comprise means for applying a triangular weighting window to each data block; and the means for reconstructing time-domain data includes means for combining the first half of each reconstructed data block with the second half of a reconstructed data block saved from processing the previous data block, time-domain samples with a uniform envelope are reconstructed and unwanted artifacts of block processing are minimized.
  • the system further comprises a noise monitor to provide an indication of when use of noise reduction would be desirable; and means for selecting the noise-reduced signal when noise level detected in the noise monitor is detected as relatively high, and for selecting the original speech with noise signal when the detected noise level is relatively low.
  • the invention may also be defined in terms of a method for reducing noise in signals received by a single microphone in a noise environment.
  • the method comprises the steps of transforming blocks of input data from a single microphone from a time-domain representation to a frequency-domain representation; filtering out selected frequency bands to minimize the effect known noise sources; detecting the presence of speech in each block of data signals; estimating noise by updating a noise spectrum estimate when no speech is detected; subtracting the noise spectrum estimate from the input speech and noise signals; and emphasizing speech signal components with respect to noise signal components, by raising the result of the subtracting step to the Nth power, where N is a positive quantity greater than one, to provide frequency-domain speech signal data with a reduced noise content.
  • the method may also include the step of reconstructing time-domain data from the noise-reduced speech signal in the frequency domain, including transforming blocks of data from the frequency domain back into the time domain, whereby the noise-reduced speech signals are more intelligible in voice communication systems.
  • the method includes the step of transmitting the noise-reduced speech signals in the frequency domain to an automatic speech recognition (ASR) system, whereby the ASR system operates more reliably to generate selected control signals.
  • ASR automatic speech recognition
  • the method step of emphasizing speech signal components includes raising signals in the frequency domain by a power N, where N is a positive quantity greater than one.
  • the method further includes the steps of presenting input signals to the noise reduction system in blocks of "A" samples each; presenting data blocks of size "2A” samples to the FFT circuit; combining input signal blocks of "A" samples in pairs to form data blocks, the combining step including using each input signal block twice, such that a currently input signal block is placed in a second half of a current data block and is then placed in a first half of a next data block; applying a triangular weighting window to each data block; and in the reconstructing step, combining the first half of each reconstructed data block with the second half of a reconstructed data block saved from processing the previous data block. Time-domain samples with a uniform envelope are reconstructed and unwanted artifacts of block processing are minimized with use of this method.
  • the method may further comprise the steps of continually monitoring the noise level with a noise monitor, to provide an indication of when use of noise reduction would be desirable; selecting the noise-reduced signal when the noise level detected by the noise monitor is detected as relatively high; and selecting the original speech and noise signal when the detected noise level is relatively low.
  • the present invention is concerned with a technique for significantly reducing the effects of noise in the detection of speech in a noisy and reverberant environment, such as the interior of a moving automobile.
  • the quality of speech transmission from mobile telephones in automobiles has long been known to be poor much of the time.
  • Noise from within and outside the vehicle result in a relatively low signal-to-noise ratio and reverberation of sounds within the vehicle further degrades the speech signals.
  • Available technologies for automatic speech recognition (ASR) and speech compression are at best degraded, and may not operate at all in the environment of the automobile.
  • a noisy speech signal is converted to digital samples and is input a block of samples at a time for processing in a fast Fourier transform (FFT) circuit, as indicated in block 10.
  • FFT fast Fourier transform
  • the signal is first bandpass filtered, as also indicated in block 10.
  • the magnitude spectrum is computed, as indicated in block 12, as the absolute value of the FFT function.
  • each block of data still in the frequency domain, is analyzed to detect the presence or absence of speech, as indicated in block 14.
  • An essential aspect of the invention is to reduce noise by spectral subtraction of noise spectrum estimate. Ideally, this estimate should be based on data obtained when speech is absent.
  • the noise spectrum estimate is not updated, but if speech is absent the noise estimate is updated.
  • the noise spectrum estimate is subtracted from the noisy speech signal spectrum, still in the frequency domain. Then, as indicated in block 20, speech is further emphasized over any residual noise by raising the speech signal (obtained after spectral subtraction of the noise) to the n th power, where n is optimized to provide the most desirable result. Finally, as indicated in block 22, the blocks of data in the frequency domain are subjected to inverse transformation by an inverse FFT circuit, which outputs a "cleaned" speech signal in the time domain.
  • FIG. 2 The functions depicted in FIG. 1 are depicted in more detail in FIG. 2.
  • the general parameter set referred to in FIG. 2 is defined in the following table: Parameter Name Description Range Units
  • a Block size (FFT size is 2A) Real positive integer (usually a power of 2) Samples B Input low cut-off point 0-parameter C Frequency (Hz) C Input high cut-off point Parameter B-sample rate/2 Frequency (Hz) D Spectral compression factor Real positive (greater than 1) Unitless E Speech location lower limit 0-parameter F Frequency (Hz) F Speech location upper limit Parameter E- sample rate/2 Frequency (Hz) G Running average energy update parameter Real positive (between 0 and 1) Unitless H Speech detect threshold parameter Real positive Unitless I Running average noise spectrum update parameter Real positive (between 0 and 1) Unitless J Speech enhancement parameter Real positive (greater than 1) Unitless
  • a Block size (FFT size is 2A) Real positive integer (usually a power of 2) Samples B Input low cut-off point
  • the functions shown in FIG. 2 may be implemented in any desired hardware or software configuration.
  • the noise cancellation system was implemented as software with code in a Microsoft Visual C++ compiler running on a personal computer in real time.
  • Input speech signals are sampled and input in blocks of A samples each.
  • Computation blocks for FFT processing are formed to contain 2A data samples each.
  • the FFT point size is 2A.
  • A may be 128 samples and 2A, 256 samples.
  • Rectangle 40 in FIG. 2 indicates the input of blocks of data.
  • Rectangle 42 indicates that each data computation block of 2A samples is formed from the stream of A-sized blocks in overlapping fashion. More specifically, if the incoming stream of A-sized blocks are designated as block (a), block (b), block (c), block (d) and so forth, then the first data computation block is formed from blocks (a) and (b) together, the next data computation block is formed from blocks (b) and (c) together, the next from blocks (c) and (d) together, and so forth.
  • the reason for overlapping the blocks in this way is to minimize sound artifacts that can be introduced by serially processing the blocks of data.
  • each data computation block is subjected to "windowing" by a triangular weighting function having the profile of an isosceles triangle centered on the data computation block.
  • a triangular weighting function having the profile of an isosceles triangle centered on the data computation block.
  • a maximum weight is applied to a sample or samples at the center of the data computation block, and progressively less weight is applied to samples towards the leading and trailing edges of the block.
  • these triangular windows also overlap.
  • the signals are later converted to the frequency domain and back to the time domain, the contributions from each adjacent pair of overlapping data computation blocks combine to produce a set of samples having a relatively uniform amplitude envelope.
  • each successive data block is formed and windowed, it is introduced to FFT processing, as indicated in rectangle 46, and then subjected to bandpass filtering between limits defined by parameters B and C, as indicated in rectangle 48.
  • This filtering step eliminates noise at very low and very high frequencies, such as below 300 Hz and above 3,850 Hz.
  • a magnitude spectrum S is computed and placed in a compressed domain using parameter D.
  • S compressed S 1/D .
  • the speech energy of the current data block is computed by summing the energy in the frequency range given by parameters E and F, such as 400 to 800 Hz, where speech is most likely to be dominant.
  • decision block 56 the current speech energy is compared with H times the average speech energy E avg , which provides a continually adapting speech detection threshold. If the current speech energy is greater that H* E avg , then the noise spectrum is not updated, as indicated by path 58.
  • the speech spectrum is then computed as the difference between the current spectrum and the noise spectrum estimate, as indicated in rectangle 62.
  • speech enhancement step 64 in which the speech spectrum, together with any residual noise component, is raised to the power J, where J is selected to be greater than one. Raising the signal to a power greater than one further distinguishes speech components from noise components.
  • the speech signals are to be transmitted to a human user of the system, they must next be transformed back to the time domain.
  • Reconstruction of the time domain waveform is also performed on a block by block basis.
  • An inverse FFT operation is performed on each data block, as indicated in rectangle 66.
  • the triangularly windowed data samples that result must be added together in a manner that will produce a uniform data envelope for the reconstructed waveform.
  • the first half of a reconstructed data block is added to the second half of the previously converted block of data, as indicated in block 68. Because these two half-blocks were originally subject to triangular windowing, they now combine in a complementary way to produce a uniform signal envelope.
  • the second half of the current block is saved for the next block iteration, as indicated in rectangle 70.
  • the combined A samples from the current and previous blocks are output, as indicated in rectangle 72.
  • a standard "star search” technique may be used, varying one parameter of the method described above while holding all others fixed. Ideally, this should be repeated for each type of speech and for different noise conditions.
  • One of the most critical parameters is the speech emphasis term, J. This was varied from 1.5 to 2.5 while testing the recognition accuracy for each setting of J. The optimum parameter value indicated was for use of the invention in the presence of freeway road and vehicle noise and for spoken connected digits data.
  • random noise indicated by graph 80
  • graph 80 has a distinctive 'spike' in its autocorrelation function 82
  • a sine wave has a periodic auto-correlation function.
  • a segment of speech 84 has strong components that are periodic sine waves. Therefore, the speech correlates strongly over several milliseconds, as indicated at 86.
  • the noise 80 correlates strongly only at the zero delay point, as indicated by the spike in its autocorrelation function 82. In the correlation domain, the spike due to noise can be easily zeroed out and this is the basis of the spectral subtraction approach used in the present invention.
  • the system of the invention has been tested under practical conditions in a moving vehicle, on a freeway with the windows closed and air-conditioning on, and also with the windows partly open.
  • Two types of microphones were considered, omni-directional and unidirectional. Not unexpectedly, the unidirectional microphone led to significantly better recognition accuracy for all background noise levels. The highest recognition accuracy obtained was 86% from freeway driving with the windows up and air conditioning on using connected digits speech data.
  • the in-vehicle data were initially collected using a digital recorder and the microphone placement was selected to maximize signal-to-noise ratio (SNR). For both the omni-directional and the unidirectional microphone the position that yields the greatest signal was just above the driver's visor (i.e., directly in front of the source). All the tests were conducted using the passenger as the point source for speech. Since the car cabin is symmetric, the results for the driver's side are expected to be equivalent to those obtained from the passenger side.
  • the speech recorded on the digital recorder in the automobile was sampled at 44.1 kHz and subsequently down-sampled to 8 kHz. In order to ensure the integrity of the audio files after down sampling, the files were tested with an automatic speech recognition (ASR) system. No degradation in ASR performance was observed for a file recorded at 44.1 kHz and down-sampled to 8 kHz.
  • ASR automatic speech recognition
  • a software package designed by Lemout and Hauspie ASR1500 was utilized for testing since it allowed for connected digits and has a relatively short response time.
  • the vocabulary tested consisted of eleven digits; 1-9, zero and oh. Connected digits were selected in order to account for the co-articulation factors in recognition process. In the test procedure, each digit is pronounced approximately fifteen times during a dialogue of a random series of connected digits.
  • the recognition accuracy for the digits is significantly improved after the removal of the background noise.
  • recognition rates improved from 47% to 86% for a unidirectional microphone, and from 16% to 78% for an omni-directional microphone.
  • recognition rates improved from 46% to 83% for a unidirectional microphone, and from less than 10% to 39% for the omni-directional microphone.
  • background noise level monitoring system 90 may be incorporated into the standard noise cancellation system of the invention, which would then operate only when a specified level of background noise is present. This would eliminate speech degradation from the processing when there is no background noise.
  • the decision need not be a "hard” (on or off) one. Rather the modified system would appropriately blend the processed and unprocessed speech in a continuously varying manner such that the effect of turning on the processing in high noise conditions would not be noticeable to the system user.
  • the monitored noise level is compared against an upper threshold, as indicated in decision block 92, and if the noise exceeds the threshold, the system selects processed (noise-reduced) speech as indicated in rectangle 94.
  • the monitored noise level is currently below the upper threshold, it is compared with a tower threshold, as indicated in decision block 96. If the noise is below the lower threshold, the original unprocessed speech is selected, as indicated in rectangle 98. If the monitored noise is between the upper and lower thresholds, the system selects a blend of inputs from the original speech and noise-reduced speech signals, as indicated in rectangle 100.
  • the noise reduction system is incorporated into an automatic speech recognition (ASR) system 104 (FIG. 5).
  • ASR automatic speech recognition
  • the noise reduction system is the same as the one illustrated in FIG. 1, but without the final inverse FFT process. This will eliminate some of the speech artifacts that are created when transforming back to the time domain waveform. Where the application calls for voice control of the ASR system only, there is no need to reconstruct the time domain waveform.
  • the inverse FFT function is eliminated from the noise cancellation system and the output of the noise cancellation system is coupled directly to frequency domain inputs 106 of the ASR system 104, which generates appropriate output control signals 108 in response to detection of input speech commands.
  • the present invention represents a significant advance in noise reduction for a single-microphone installed in noisy environment, such as a moving automobile.
  • the invention provides a "cleaned” or noise-reduced speech signal that is more intelligible to the human ear and improves reliability of ASR systems.
  • the system of the invention produces either time-domain output for transmission over voice communication systems, or frequency-domain output for direct connection to an ASR system.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Quality & Reliability (AREA)
  • Signal Processing (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Circuit For Audible Band Transducer (AREA)
  • Fittings On The Vehicle Exterior For Carrying Loads, And Devices For Holding Or Mounting Articles (AREA)
EP00118147A 1999-09-01 2000-08-29 Rauschverminderungsverfahren in einem Sprachsignal mit einem einzigen Mikrophon Withdrawn EP1081685A3 (de)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US388266 1989-08-01
US38826699A 1999-09-01 1999-09-01

Publications (2)

Publication Number Publication Date
EP1081685A2 true EP1081685A2 (de) 2001-03-07
EP1081685A3 EP1081685A3 (de) 2002-04-24

Family

ID=23533388

Family Applications (1)

Application Number Title Priority Date Filing Date
EP00118147A Withdrawn EP1081685A3 (de) 1999-09-01 2000-08-29 Rauschverminderungsverfahren in einem Sprachsignal mit einem einzigen Mikrophon

Country Status (2)

Country Link
EP (1) EP1081685A3 (de)
JP (1) JP2001092491A (de)

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100421013B1 (ko) * 2001-08-10 2004-03-04 삼성전자주식회사 음성 향상 시스템 및 방법
GB2437559A (en) * 2006-04-26 2007-10-31 Zarlink Semiconductor Inc System for reducing background noise in a speech signal by use of a fast Fourier transform
CN101320566B (zh) * 2008-06-30 2010-10-20 中国人民解放军第四军医大学 基于多带谱减法的非空气传导语音增强方法
CN102930870A (zh) * 2012-09-27 2013-02-13 福州大学 利用抗噪幂归一化倒谱系数的鸟类声音识别方法
US8538749B2 (en) 2008-07-18 2013-09-17 Qualcomm Incorporated Systems, methods, apparatus, and computer program products for enhanced intelligibility
US8615393B2 (en) 2006-11-15 2013-12-24 Microsoft Corporation Noise suppressor for speech recognition
US8831936B2 (en) 2008-05-29 2014-09-09 Qualcomm Incorporated Systems, methods, apparatus, and computer program products for speech signal processing using spectral contrast enhancement
US9053697B2 (en) 2010-06-01 2015-06-09 Qualcomm Incorporated Systems, methods, devices, apparatus, and computer program products for audio equalization
JP2015169915A (ja) * 2014-03-10 2015-09-28 公立大学法人広島市立大学 アクティブノイズ制御装置およびアクティブノイズ制御方法
CN104978955A (zh) * 2014-04-14 2015-10-14 美的集团股份有限公司 语音控制方法和***
US9202456B2 (en) 2009-04-23 2015-12-01 Qualcomm Incorporated Systems, methods, apparatus, and computer-readable media for automatic control of active noise cancellation
WO2016094418A1 (en) * 2014-12-09 2016-06-16 Knowles Electronics, Llc Dynamic local asr vocabulary
US9536540B2 (en) 2013-07-19 2017-01-03 Knowles Electronics, Llc Speech signal separation and synthesis based on auditory scene analysis and speech modeling
US9820042B1 (en) 2016-05-02 2017-11-14 Knowles Electronics, Llc Stereo separation and directional suppression with omni-directional microphones
US9838784B2 (en) 2009-12-02 2017-12-05 Knowles Electronics, Llc Directional audio capture
US9978388B2 (en) 2014-09-12 2018-05-22 Knowles Electronics, Llc Systems and methods for restoration of speech components
WO2018140020A1 (en) * 2017-01-26 2018-08-02 Nuance Communications, Inc. Methods and apparatus for asr with embedded noise reduction
US10045140B2 (en) 2015-01-07 2018-08-07 Knowles Electronics, Llc Utilizing digital microphones for low power keyword detection and noise suppression
CN111724805A (zh) * 2020-06-29 2020-09-29 北京百度网讯科技有限公司 用于处理信息的方法和装置
CN114650484A (zh) * 2022-05-23 2022-06-21 东莞市云仕电子有限公司 具有自动降噪功能的无线耳机及其使用方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
FI19992453A (fi) * 1999-11-15 2001-05-16 Nokia Mobile Phones Ltd Kohinanvaimennus
JP7231181B2 (ja) * 2018-07-17 2023-03-01 国立研究開発法人情報通信研究機構 耐雑音音声認識装置及び方法、並びにコンピュータプログラム
JPWO2023100374A1 (de) * 2021-12-03 2023-06-08

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5012519A (en) * 1987-12-25 1991-04-30 The Dsp Group, Inc. Noise reduction system
EP0637012A2 (de) * 1990-01-18 1995-02-01 Matsushita Electric Industrial Co., Ltd. Vorrichtung zur Rauschreduzierung
US5742927A (en) * 1993-02-12 1998-04-21 British Telecommunications Public Limited Company Noise reduction apparatus using spectral subtraction or scaling and signal attenuation between formant regions

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5012519A (en) * 1987-12-25 1991-04-30 The Dsp Group, Inc. Noise reduction system
EP0637012A2 (de) * 1990-01-18 1995-02-01 Matsushita Electric Industrial Co., Ltd. Vorrichtung zur Rauschreduzierung
US6038532A (en) * 1990-01-18 2000-03-14 Matsushita Electric Industrial Co., Ltd. Signal processing device for cancelling noise in a signal
US5742927A (en) * 1993-02-12 1998-04-21 British Telecommunications Public Limited Company Noise reduction apparatus using spectral subtraction or scaling and signal attenuation between formant regions

Cited By (27)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100421013B1 (ko) * 2001-08-10 2004-03-04 삼성전자주식회사 음성 향상 시스템 및 방법
GB2437559A (en) * 2006-04-26 2007-10-31 Zarlink Semiconductor Inc System for reducing background noise in a speech signal by use of a fast Fourier transform
GB2437559B (en) * 2006-04-26 2010-12-22 Zarlink Semiconductor Inc Low complexity noise reduction method
US8615393B2 (en) 2006-11-15 2013-12-24 Microsoft Corporation Noise suppressor for speech recognition
US8831936B2 (en) 2008-05-29 2014-09-09 Qualcomm Incorporated Systems, methods, apparatus, and computer program products for speech signal processing using spectral contrast enhancement
CN101320566B (zh) * 2008-06-30 2010-10-20 中国人民解放军第四军医大学 基于多带谱减法的非空气传导语音增强方法
US8538749B2 (en) 2008-07-18 2013-09-17 Qualcomm Incorporated Systems, methods, apparatus, and computer program products for enhanced intelligibility
US9202456B2 (en) 2009-04-23 2015-12-01 Qualcomm Incorporated Systems, methods, apparatus, and computer-readable media for automatic control of active noise cancellation
US9838784B2 (en) 2009-12-02 2017-12-05 Knowles Electronics, Llc Directional audio capture
US9053697B2 (en) 2010-06-01 2015-06-09 Qualcomm Incorporated Systems, methods, devices, apparatus, and computer program products for audio equalization
CN102930870A (zh) * 2012-09-27 2013-02-13 福州大学 利用抗噪幂归一化倒谱系数的鸟类声音识别方法
CN102930870B (zh) * 2012-09-27 2014-04-09 福州大学 利用抗噪幂归一化倒谱系数的鸟类声音识别方法
US9536540B2 (en) 2013-07-19 2017-01-03 Knowles Electronics, Llc Speech signal separation and synthesis based on auditory scene analysis and speech modeling
JP2015169915A (ja) * 2014-03-10 2015-09-28 公立大学法人広島市立大学 アクティブノイズ制御装置およびアクティブノイズ制御方法
CN104978955A (zh) * 2014-04-14 2015-10-14 美的集团股份有限公司 语音控制方法和***
US9978388B2 (en) 2014-09-12 2018-05-22 Knowles Electronics, Llc Systems and methods for restoration of speech components
WO2016094418A1 (en) * 2014-12-09 2016-06-16 Knowles Electronics, Llc Dynamic local asr vocabulary
US10045140B2 (en) 2015-01-07 2018-08-07 Knowles Electronics, Llc Utilizing digital microphones for low power keyword detection and noise suppression
US9820042B1 (en) 2016-05-02 2017-11-14 Knowles Electronics, Llc Stereo separation and directional suppression with omni-directional microphones
WO2018140020A1 (en) * 2017-01-26 2018-08-02 Nuance Communications, Inc. Methods and apparatus for asr with embedded noise reduction
CN110268471A (zh) * 2017-01-26 2019-09-20 诺昂世通讯公司 具有嵌入式降噪的asr的方法和设备
EP3574499A4 (de) * 2017-01-26 2020-09-09 Nuance Communications, Inc. Verfahren und vorrichtung für asr mit eingebetteter rauschminderung
US11308946B2 (en) 2017-01-26 2022-04-19 Cerence Operating Company Methods and apparatus for ASR with embedded noise reduction
CN110268471B (zh) * 2017-01-26 2023-05-02 赛伦斯运营公司 具有嵌入式降噪的asr的方法和设备
CN111724805A (zh) * 2020-06-29 2020-09-29 北京百度网讯科技有限公司 用于处理信息的方法和装置
CN114650484A (zh) * 2022-05-23 2022-06-21 东莞市云仕电子有限公司 具有自动降噪功能的无线耳机及其使用方法
CN114650484B (zh) * 2022-05-23 2022-09-06 东莞市云仕电子有限公司 具有自动降噪功能的无线耳机及其使用方法

Also Published As

Publication number Publication date
EP1081685A3 (de) 2002-04-24
JP2001092491A (ja) 2001-04-06

Similar Documents

Publication Publication Date Title
EP1081685A2 (de) Rauschverminderungsverfahren in einem Sprachsignal mit einem einzigen Mikrophon
EP1739657B1 (de) Sprachsignalverbesserung
US8010355B2 (en) Low complexity noise reduction method
US6487257B1 (en) Signal noise reduction by time-domain spectral subtraction using fixed filters
US8249861B2 (en) High frequency compression integration
EP1080465B1 (de) Rauschunterdrückung mittels spektraler subtraktion unter verwendung von linearem faltungsprodukt und kausaler filterung
KR100851716B1 (ko) 바크 대역 위너 필터링 및 변형된 도블링거 잡음 추정에기반한 잡음 억제
EP2244254B1 (de) Gegen hohe Anregungsgeräusche unempfindliches System zum Ausgleich von Umgebungsgeräuschen
EP1855456B1 (de) Echoverringerung für zeitvariante Systeme
Yang Frequency domain noise suppression approaches in mobile telephone systems
EP2416315B1 (de) Rauschunterdrückungseinrichtung
US5878389A (en) Method and system for generating an estimated clean speech signal from a noisy speech signal
US20060222184A1 (en) Multi-channel adaptive speech signal processing system with noise reduction
US6510224B1 (en) Enhancement of near-end voice signals in an echo suppression system
KR20070085729A (ko) 바크 밴드 위너 필터 및 선형 감쇠를 이용한 노이즈 감소및 컴포트 노이즈 이득 제어
US20140244245A1 (en) Method for soundproofing an audio signal by an algorithm with a variable spectral gain and a dynamically modulatable hardness
KR100470523B1 (ko) 마이크로폰 신호로부터 스피커 간섭을 제거하기 위한 필터 시스템
US7917359B2 (en) Noise suppressor for removing irregular noise
Itoh et al. Environmental noise reduction based on speech/non-speech identification for hearing aids
EP2490218B1 (de) Verfahren zur Interferenzunterdrückung
US6507623B1 (en) Signal noise reduction by time-domain spectral subtraction
US20060184361A1 (en) Method and apparatus for reducing an interference noise signal fraction in a microphone signal
Esch et al. Combined reduction of time varying harmonic and stationary noise using frequency warping
US11227622B2 (en) Speech communication system and method for improving speech intelligibility
JP2003517761A (ja) 通信システムにおける音響バックグラウンドノイズを抑制するための方法と装置

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

AK Designated contracting states

Kind code of ref document: A2

Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LI LU MC NL PT SE

AX Request for extension of the european patent

Free format text: AL;LT;LV;MK;RO;SI

PUAL Search report despatched

Free format text: ORIGINAL CODE: 0009013

AK Designated contracting states

Kind code of ref document: A3

Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LI LU MC NL PT SE

AX Request for extension of the european patent

Free format text: AL;LT;LV;MK;RO;SI

17P Request for examination filed

Effective date: 20020712

AKX Designation fees paid

Free format text: AT BE CH CY DE DK ES FI FR GB GR IE IT LI LU MC NL PT SE

RAP1 Party data changed (applicant data changed or rights of an application transferred)

Owner name: NORTHROP GRUMMAN CORPORATION

RAP1 Party data changed (applicant data changed or rights of an application transferred)

Owner name: NORTHROP GRUMMAN CORPORATION

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE APPLICATION IS DEEMED TO BE WITHDRAWN

18D Application deemed to be withdrawn

Effective date: 20050301