Yule walker based low-complexity voice activity detector in noise suppression systems
원문보기
IPC분류정보
국가/구분
United States(US) Patent
등록
국제특허분류(IPC7판)
G10L-025/00
G10L-021/00
G10L-021/02
G10L-015/00
출원번호
US-0890268
(2007-08-03)
등록번호
US-8775168
(2014-07-08)
발명자
/ 주소
Muralidhar, Karthik
Krishna, Anoop Kumar
출원인 / 주소
STMicroelectronics Asia Pacific PTE, Ltd.
대리인 / 주소
Munck Wilson Mandala, LLP
인용정보
피인용 횟수 :
1인용 특허 :
15
초록▼
A Yule-Walker based, low-complexity voice activity detector (VAD) is disclosed. An input signal is typically noisy speech (i.e., corrupted with, for example, babble noise). In one embodiment, a first initialization stage of the VAD computes an occurrence of a silent period within the input signal an
A Yule-Walker based, low-complexity voice activity detector (VAD) is disclosed. An input signal is typically noisy speech (i.e., corrupted with, for example, babble noise). In one embodiment, a first initialization stage of the VAD computes an occurrence of a silent period within the input signal and the AR parameters. The VAD could accordingly compute a tentative adaptive threshold and output hypothesis H1 (which means speech is present) during this stage. During the second initialization stage, the VAD generally builds a database of associated values and computes the adaptive threshold accordingly. The second initialization stage could also output tentative VAD decisions based on the tentative threshold computed in the first initialization stage. Finally, the VAD periodically retrains or updates AR parameters, threshold values and/or the database and outputs VAD decisions accordingly.
대표청구항▼
1. A method of detecting voice activity from an input signal having an initial silent period and a speech period, the method comprising: determining occurrence of the initial silent period;computing an autoregressive (AR) parameter from the initial silent period;storing a test statistic of the input
1. A method of detecting voice activity from an input signal having an initial silent period and a speech period, the method comprising: determining occurrence of the initial silent period;computing an autoregressive (AR) parameter from the initial silent period;storing a test statistic of the input signal calculated during both the initial silent and speech periods in a database, wherein the test statistic is calculated from a product of the AR parameter adjusted by an autocorrelation function with a transpose of the AR parameter adjusted by the autocorrelation function;computing a threshold from the database based on one or more values of the test statistic during the initial silent period and one or more values of the test statistic during the speech period; andoutputting a decision value based on the AR parameter and the threshold. 2. The method of claim 1, wherein computing the AR parameter further comprises using a Yule-Walker relationship. 3. The method of claim 1, wherein computing the threshold further comprises computing an adaptive threshold using at least one of: the AR parameters and the database. 4. The method of claim 3, further comprising: periodically updating at least one of: the AR parameter, the adaptive threshold value and the database. 5. The method of claim 3, further comprising: periodically updating the adaptive threshold value at least once between one of two silent periods separated by a speech period and two speech periods separated by a silent period. 6. The method of claim 3, wherein outputting the decision value further comprises outputting the decision value based on the adaptive threshold. 7. The method of claim 1, wherein the test statistic Ψk is calculated using: Ψk=[Ran−r]T[Ran−r]where R is an AR correlation matrix, an are coefficients of an infinite impulse response (IIR) filter, and r is and autocorrelation function (ACF) correlation matrix. 8. The method of claim 1, further comprising: computing a tentative adaptive threshold from the initial silent period. 9. The method of claim 1, further comprising: periodically updating the AR parameter when a second silent period has a duration greater than or equal to 30 blocks. 10. The method of claim 1, wherein the database comprises a logarithm of a smoothed local maxima of a test statistic of the input signal computed on a block by block basis. 11. A voice activity detector (VAD), comprising: an input configured to receive a signal having an initial silent period and a speech period;a first circuit configured to determine occurrence of the initial silent period,compute an autoregressive (AR) parameter from the initial silent period, andcompute a threshold based on one or more values of a test statistic during the initial silent period and one or more values of the test statistic during the speech period, wherein the test statistic is calculated from a product of the AR parameter adjusted by an autocorrelation function with a transpose of the AR parameter adjusted by an autocorrelation function;a memory configured to store a database of a test statistic of the input signal calculated during both the silent and speech periods; anda second circuit configured to output a decision value based on the AR parameter the threshold calculated in the first circuit. 12. The VAD of claim 11, wherein the first circuit is configured to compute the AR parameter using a Yule-Walker relationship. 13. The VAD of claim 11, wherein the second circuit is configured to compute an adaptive threshold using at least one of: the AR parameters and the database. 14. The VAD of claim 13, further comprising a third circuit configured to output a decision value based on at least one of: the AR parameter, the threshold and the database. 15. The VAD of claim 13, wherein the second circuit is configured to build the database of the test statistic of the input signal. 16. The VAD of claim 15, wherein the test statistic Ψk is computed using: Ψk=[Ran−r]T[Ran−r]where R is an AR correlation matrix, an are coefficients of an infinite impulse response (IIR) filter, and r is and autocorrelation function (ACF) correlation matrix. 17. The VAD of claim 13, further comprising a third circuit configured to periodically update the adaptive threshold value at least once between one of two silent periods separated by a speech period and two speech periods separated by a silent period. 18. The VAD of claim 11, wherein the first circuit is configured to compute a tentative adaptive threshold from the silent period. 19. The VAD of claim 11, further comprising a third circuit configured to periodically update the AR parameter when a second silent period has a duration greater than or equal to 30 blocks. 20. A method of using a voice activity detector (VAD), the method comprising: receiving an input signal having an initial silent period and a speech period;computing an autoregressive (AR) parameter from the initial silent period using a Yule-Walker relationship;storing a test statistic of the input signal calculated during both the initial silent period and the speech period in a database;computing an adaptive threshold based on one or more values of the test statistic during the silent period and one or more values of the test statistic during the speech period wherein the test statistic is calculated from a product of the AR parameter adjusted by an autocorrelation function with a transpose of the AR parameter adjusted by an autocorrelation function; andoutputting a decision value based on the AR parameter and the adaptive threshold. 21. The method of claim 19, further comprising: periodically updating at least one of: the AR parameter, the adaptive threshold value and the database.
연구과제 타임라인
LOADING...
LOADING...
LOADING...
LOADING...
LOADING...
이 특허에 인용된 특허 (15)
Li,Dunling, Background noise estimation method for an improved G.729 annex B compliant voice activity detection circuit.
Bhattacharya, Puranjoy; Singhal, Manoj Kumar; Dummy, Sangeetha, Preprocessing modules for quality enhancement of MBE coders and decoders for signals having transmission path characteristics.
Bou Ghazale,Sahar E.; Asadi,Ayman O.; Assaleh,Khaled, System and method for a endpoint detection of speech for improved speech recognition in noisy environments.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.