Apparatus and method for voice activity detection

• US 8,442,817 B2
• Filed: 12/23/2004
• Issued: 05/14/2013
• Est. Priority Date: 12/25/2003
• Status: Active Grant

• Current Assignee: NTT Docomo Incorporated (Nippon Telegraph and Telephone Corporation)
• Original Assignee: NTT Docomo Incorporated (Nippon Telegraph and Telephone Corporation)
• Inventors: Naka, Nobuhiko, Ohya, Tomoyuki
• Primary Examiner(s): Desir, Pierre-Louis
• Assistant Examiner(s): BAKER, MATTHEW H
• Application Number:
  US11/024,267
• Publication Number:
  US 20050154583A1
• Time in Patent Office: 3,064 Days
• Field of Search: 704211-217
• US Class Current: 704/217
• CPC Class Codes: G10L 25/06 the extracted parameters being correlation coefficients
• 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

Abstract

It is provided a voice activity decision apparatus capable of accurately performing the decision on the state being associated with a sound interval or a silence interval also in terms of the input signal having many aperiodic components and/or plural mixed different periodic components. The apparatus 1 comprises: an autocorrelation calculating unit 11 for calculating autocorrelation values of an input signal; a delay calculating unit 12 for calculating plural delays at which autocorrelation values calculated by the autocorrelation calculating unit 11 become maximums; a noise deciding unit 13 for deciding whether the input signal is a noise or not based on the plurality of delays calculated by the delay calculating unit 12; and an activity decision unit 14 for performing the activity decision in terms of the input signal based on results of decision by the noise deciding unit 13 and the input signal.

Claims

1. A voice activity decision apparatus comprising:

a processor in communication with a memory, wherein the processor is configured to receive an input signal;

an autocorrelation calculation module stored in the memory and executable with the processor, the autocorrelation calculation module configured to calculate a plurality of autocorrelation values for the input signal, the plurality of autocorrelation values calculated within a predetermined interval;

a delay calculation module stored in the memory and executable with the processor, the delay calculation module configured to receive the autocorrelation values calculated within the predetermined interval by the autocorrelation calculation module, and further configured to identify local maximum valued autocorrelation values within the autocorrelation values, and the delay calculation module further configured to calculate a plurality of delays within the predetermined interval, wherein the delays comprise a respective delay for each of the local maximum valued autocorrelation values;

a noise decision module stored in the memory and executable with the processor, the noise decision module configured to receive the delays, the noise decision module further configured to determine whether variations between the received delays are less than a threshold for at least a predetermined period of time, and further configured to generate a signal characteristic determination that the input signal includes a non-noise portion based upon determination that the variations between the received delays are less than the threshold for the at least the predetermined period of time;

an activity detector module stored in the memory and executable with the processor, the activity detector module configured to receive the signal characteristic determination of the input signal, and further configured to determine a signal activity decision based on the signal characteristic determination; and

a noise estimation module stored in the memory and executable with the processor, the noise estimation module configured to receive the input signal and generate a noise estimate for the input signal,

wherein the activity detector module is further configured to determine the signal activity decision based on the signal characteristic determination, the input signal, and the noise estimate, and the noise estimation module is further configured to adapt the noise estimate based on the signal activity decision.

2. The system of claim 1, wherein the activity detector module is further configured to determine the signal activity decision based on the input signal and the signal characteristic determination.

3. The system of claim 1, wherein the activity detector module is further configured to receive the input signal and to determine the signal activity decision based on a signal analysis of the input signal and the signal characteristic determination

4. The system of claim 3, wherein the signal analysis of the input signal comprises a signal measurement comprising at least one of a power measurement, a spectrum envelope measurement, a zero-crossing analysis, or a combination thereof.

5. The system of claim 1, wherein the delay calculation module is further configured to calculate the respective delay for each of the local maximum valued autocorrelation values in an order, wherein the order is determined with the delay calculation module based on a magnitude of each of the local maximum valued autocorrelation values.

6. The system of claim 1, wherein the delay calculation module is further configured to divide a delay-observation interval into delay intervals, and

the delay calculation module further configured to identify a maximum valued autocorrelation value within each of the delay intervals as one of the local maximum valued autocorrelation values.

7. The system of claim 6, wherein the delay calculation module is further configured to divide the delay-observation interval into a contiguous series of the delay intervals, wherein each successive delay interval in the contiguous series of the delay intervals is longer than the one of the delay intervals that precedes the successive delay interval by a predetermined amount of delay.

8. The system of claim 6, wherein the delay calculation module is further configured to divide the delay-observation interval into a contiguous series of delay intervals, wherein each successive delay interval in the contiguous series of delay intervals is twice as long as the one of the delay intervals that precedes the successive delay interval.

9. The system of claim 6, wherein the delay calculation module is further configured to divide the delay-observation interval into a contiguous series of delay intervals, wherein the delay intervals are of substantially uniform size.

10. The system of claim 1, wherein the signal activity decision is indicative that the input signal is one of noise or speech.

11. The system of claim 1, wherein the noise decision module is further configured to calculate variations between the received delays.

12. The system of claim 11, wherein each of the calculated variations is a difference between the delay of each of the local maximum valued autocorrelation values and the delay of an adjacent local maximum auto correlation value.

13. The system of claim 1, further comprising a noise estimation module configured to adapt a noise estimate so as to generate a lower value of the noise estimate when the activity detector determines that the input signal is in a sound-present state, than when the activity detector determines that the input signal is in a silent state.

14. A non-transitory computer readable storage device for storing a voice activity detection program, the computer readable storage device comprising:

computer program code embodied on said computer readable storage device, wherein the computer program code is executable with a processor, and wherein the computer program code comprises;

computer program code to calculate a plurality of autocorrelation values of an input signal within a predetermined interval;

computer program code to identify local maximum autocorrelation values within the autocorrelation values calculated within the predetermined interval;

computer program code to calculate a delay for each of the local maximum autocorrelation values identified within the predetermined interval to generate a plurality of delays associated with the local maximum autocorrelations values;

computer program code to determine whether variations between the delays associated with the local maximum autocorrelation values are less than a threshold for a predetermined period of time;

computer program code to, in response to determination that the variations between the delays associated with the local maximum autocorrelation values are less than the threshold for the predetermined period of time, generate a signal characteristic determination that the input signal includes a signal component other than noise;

computer program code to determine a signal activity decision based on the signal characteristic determination;

computer program code to generate a noise estimate, wherein the computer program code to determine the signal activity decision further comprises computer program code to generate the signal activity decision based on the input signal and the noise estimate; and

computer program code to adapt the noise estimate in response to the signal activity decision.

15. The non-transitory computer readable storage device of claim 14, wherein the computer program code to determine the signal activity decision further comprises computer program code to generate the signal activity decision based on both the signal characteristic determination and the input signal.

16. The non-transitory computer readable storage device of claim 14, wherein the computer program code to determine the signal activity decision further comprises computer program code to generate the signal activity decision based on the signal characteristic determination, the input signal, and a signal analysis of the input signal.

17. The non-transitory computer readable storage device of claim 14, wherein the computer program code further comprises:

computer program code to generate a signal measurement of the input signal comprising at least one of a power measurement, a spectrum envelope measurement, a zero-crossing analysis, or a combination thereof; and

wherein the computer program code to determine the signal activity decision further comprises computer program code to generate the signal activity decision based on the signal measurement of the input signal, the input signal, and the signal characteristic determination.

18. The non-transitory computer readable storage device of claim 14, wherein the computer program code to generate the noise estimate further comprises computer program code to adjust a present value of the noise estimate based on a combination of a portion of a previous value of the noise estimate and a portion of a current value of the input signal.

19. The non-transitory computer readable storage device of claim 14, wherein the computer program code further comprises:

computer program code to generate a signal to noise ratio based upon a noise estimate of the input signal and the input signal;

computer program code to detect a sound-present state as a function of the signal to noise ratio being greater than a threshold value; and

computer program code to detect a sound-silent state as a function of the signal to noise ratio being equal to or less than the threshold value.

20. The non-transitory computer readable storage device of claim 19, wherein the computer program code further comprises:

computer program code to reduce the threshold value in response to detection of the sound-present state; and

computer program code to increase the threshold value in response to detection of the sound-silent state.

21. The non-transitory computer readable storage device of claim 14, wherein the computer program code further comprises:

computer program code to divide a delay-observation interval into delay intervals; and

wherein the computer program code to identify the local maximum autocorrelation values further comprises computer program code to identify a maximum valued autocorrelation value within each of the delay intervals; and

wherein the computer program code to calculate the delay for each of the local maximum autocorrelation values further comprises computer program code to calculate the delay for the maximum valued autocorrelation value within each of the delay intervals.

22. The non-transitory computer readable storage device of claim 21, wherein the computer program code to divide the delay-observation interval into the delay intervals further comprises:

computer program code to divide the delay-observation interval into delay intervals of substantially uniform size.

23. The non-transitory computer readable storage device of claim 21, wherein the computer program code to divide the delay-observation interval into the delay intervals further comprises:

computer program code to divide the delay-observation interval into a contiguous series of delay intervals, wherein each successive one of the delay intervals is longer by a predetermined amount.

24. The non-transitory computer readable storage device of claim 23, wherein the contiguous series of delay intervals has a predetermined length of delay, and wherein each successive one of the delay intervals is a factor of two longer.

25. The non-transitory computer readable storage device of claim 14, further comprising:

computer program code to calculate variations between the delays associated with each of the local maximum autocorrelation values.

26. The computer readable storage device of claim 14, further comprising computer program code to estimate a noise from the input signal, the noise estimated to be a first value when the signal activity decision is that the input signal is in a sound-present state, and the noise estimated to be a second value when the signal activity decision is that the input signal is in a silent state, the first value being lower than the second value.

27. A method for voice activity detection comprising:

calculating with a processor a plurality of autocorrelation values of an input signal, the autocorrelation values calculated within a predetermined interval;

identifying local maximum autocorrelation values within the autocorrelation values calculated within the predetermined interval with the processor;

calculating a delay for each of the local maximum autocorrelation values identified within the predetermined interval with the processor;

generating a plurality of delays associated with the local maximum autocorrelations values with the processor;

determining with the processor whether variations between the delays associated with the local maximum autocorrelation values are less than a threshold for a predetermined period of time;

generating an input signal characteristic determination of the input signal with the processor when determination that the variations between the delays associated with the local maximum autocorrelation values are less than the threshold for the predetermined period of time, the input signal characteristic determination indicative that the input signal includes a signal component other than noise;

generating a noise estimate of the input signal with the processor;

the processor adapting the noise estimate based on a previous signal activity decision; and

the processor determining a signal activity decision based on the input signal characteristic determination and consideration of the noise estimate of the input signal.

28. The method of claim 27, wherein generating the input signal characteristic of the input signal further comprises:

comparing each of the variations to a delay difference threshold value;

detecting that an input signal characteristic of the input signal is noise in response to at least one of the variations being greater than the delay difference threshold value; and

detecting that the input signal characteristic of the input signal is a voice signal in response to all of the variations being less than or equal to the delay difference threshold value.

29. The method of claim 27, further comprising:

generating a signal analysis of the input signal with the processor, wherein determination of the signal activity decision further includes consideration of the signal analysis of the input signal.

30. The method of claim 29, further comprising:

analyzing the input signal with the processor to generate a signal characteristic measurement associated with the input signal, the analysis comprising at least one of a power measurement, a spectrum envelope measurement, a zero-crossing analysis, or a combination thereof, wherein determination of the signal activity decision further includes consideration of the signal characteristic measurement.

31. The method of claim 27, further comprising:

generating a signal to noise ratio with the processor based upon a noise estimate of the input signal and the input signal;

detecting a sound-present state with the processor based upon the signal to noise ratio being equal to or greater than a threshold value; and

detecting a sound-silent state with the processor based upon the signal to noise ratio being less than the threshold value.

32. The method of claim 31, further comprising:

reducing the threshold value with the processor based upon detection of the sound-present state; and

increasing the threshold value with the processor based upon detection of the sound-silent state.

33. The method of claim 27, wherein a delay interval of each of the delays is substantially uniform.

34. The method of claim 27, further comprising

estimating a noise from the input signal with the processor, the noise characteristic of the input signal estimated to be a first value when the signal activity decision determines that the input signal is a sound-present state, and the noise characteristic of the input signal estimated to be a second value when the signal activity decision determines that the input signal is in a silent state, the first value being lower than the second value.