Voice activity detector

• US 6,061,647 A
• Filed: 04/30/1998
• Issued: 05/09/2000
• Est. Priority Date: 09/14/1993
• Status: Expired due to Term

• Current Assignee: LG Electronics, Inc. (LG Corporation)
• Original Assignee: British Telecommunications PLC (BT Group PLC)
• Inventors: Barrett, Paul Alexander
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): Lerner, Martin
• Application Number:
  US09/070,151
• Time in Patent Office: 740 Days
• Field of Search: 704/214, 704/215, 704/219, 704/233, 704/208, 704/210, 704/226, 704/227, 704/228
• US Class Current: 704/208
• CPC Class Codes: G10L 2025/786 Adaptive threshold
• CPC Class Codes: G10L 25/78 Detection of presence or absence of voice signals switching of direction of transmission by voice frequency in two-way loud-speaking telephone systems H04M9/10
• CPC Class Codes: H04M 19/08 with current supply sources at the substations generating ringing current H04M19/04
• CPC Class Codes: H04Q 1/46 comprising means for distinguishing between a signalling current of predetermined frequency and a complex current containing that frequency, e.g. speech current

Abstract

Speech is distinguished from noise by a spectral comparison of an input signal with a stored noise estimate. Updating of the noise estimate stored in a buffer is permitted during periods when speech is absent under control of an auxiliary detector. In order to improve operation in the presence of signals with strong harmonic components, e.g., signaling tones, an LPC prediction gain is computed from the input (x(i)) and a residual (y(i)) obtained from the input following filtering by a filter having a response complementary to the frequency spectrum of the input, and if the gain exceeds a threshold, buffer updating is suppressed.

Claims

1. A voice activity detector for detecting the presence of speech in an input signal which comprises a noise component, said detector comprising:

(a) means for storing an estimate of the noise component of an input signal;

(b) means for recognizing a spectral similarity of the input signal and the stored estimate to produce an output decision signal;

(c) means for updating the stored estimate;

(d) an auxiliary detector arranged to indicate the absence of speech from the input signal and control the means for updating so that updating occurs only when speech is indicated by the auxiliary detector to be absent from the input signal;

(e) means operable to calculate a measure of the prediction gain for the input signal;

(f) modifying means arranged to suppress said updating of the stored estimate in the event that the prediction gain exceeds a threshold value.

2. A voice activity detector according to claim 1 in which the auxiliary detector includes a voiced speech detector responsive to signals derived from a Linear Predictive Coding residual signal.

3. A voice activity detector according to claim 1 in which the measure of the prediction gain represents the prediction gain of a Linear Predictive Coding prediction of 6th order or less.

4. A voice activity detector according to claim 3 in which the measure of the prediction gain represents the prediction gain of a Linear Predictive Coding prediction of 4th order or less.

5. A voice activity detector according to claim 1 further including means for detecting one or more primary frequency components of the input signal, and means operable to compare the frequencies with a predetermined threshold and to permit suppression of said updating of the stored estimate only when said primary frequency component exceeds the threshold.

6. A voice activity detector according to claim 1 in which the prediction gain calculation means is preceded by a filter to remove an upper portion of the frequency range of the input signal, the measure calculation being performed only for frequency components lying within the passband of the filter.

7. A voice activity detector according to claim 6 having two filters having respective passbands and respective measure calculation means to calculate a measure of the prediction gain for the respective passbands, the modifying means being arranged to suppress the updating of the stored estimate in the event that the prediction gain within either passband exceeds a threshold value.

8. A voice activity detector according to claim 6 including means for subsampling a signal produced by the filter.

9. A voice activity detector according to claim 1 wherein said input signal includes signaling tones and the prediction gain exceeding a threshold value indicates the detection of signaling tones in the input signal and causes the modifying means to suppress the updating of the stored estimate.

10. A method for detecting the presence of speech in an input signal which includes a noise component, said method comprising:

(a) storing an estimate of the noise component of an input signal;

(b) recognizing a spectral similarity of the input signal and the stored estimate to produce an output decision signal;

(c) updating the stored estimate;

(d) using an auxiliary detector arranged to indicate the absence of speech from the input signal and control the updating step so that updating occurs only when speech is indicated by the auxiliary detector to be absent from the input signal;

(e) generating a measure of the prediction gain for the input signal;

(f) suppressing said updating of the stored estimate in the event that the prediction gain exceeds a threshold value.

11. A method as in claim 10 in which the auxiliary detector uses a voiced speech detector responsive to signals derived from a Linear Predictive Coding residual signal.

12. A method as in claim 10 in which the measure of the prediction gain represents the prediction gain of a Linear Predictive coding prediction of 6^th order or less.

13. A method as in claim 12 in which the measure of the prediction gain represents the prediction gain of a Linear Predictive Coding prediction of 4^th order or less.

14. A method as in claim 10 further including:

detecting one or more primary frequency components of the input signal, and

comparing the frequencies with a predetermined threshold and to permit suppression of said updating of the stored estimate only when said primary frequency component exceeds the threshold.

15. A method as in claim 10 in which the calculation of the measure of the prediction gain is preceded by filtered removal of an upper portion of the frequency range of the input signal, the measure calculation being performed only for frequency components lying within the passband of the filter.

16. A method as in claim 15 wherein two filters are used having respective passbands and respective measure calculation means to calculate measures of the prediction gain for the respective passbands, the modifying step suppressing the updating of the stored estimate in the event that the prediction gain within either passband exceeds a threshold value.

17. A method as in claim 15 including subsampling a signal produced by the filter.

18. A method as in claim 10 wherein said input signal includes signaling tones and the prediction gain exceeding a threshold value indicates the detection of signaling tones in the input signal and causes the modifying step to suppress the updating of the stored estimate.