Apparatus and method for recognizing voice with reduced sensitivity to ambient noise

• US 6,411,928 B2
• Filed: 07/21/1997
• Issued: 06/25/2002
• Est. Priority Date: 02/09/1990
• Status: Expired due to Fees

• Current Assignee: Sanyo Electric Company Limited (Panasonic Holdings Corporation)
• Original Assignee: Sanyo Electric Company Limited (Panasonic Holdings Corporation)
• Inventors: Suzuki, Ryuji, Iida, Masayuki, Tsurufuji, Shin-ichi
• Primary Examiner(s): EDOUARD, PATRICK NESTOR
• Application Number:
  US08/897,734
• Publication Number:
  US 20010029449A1
• Time in Patent Office: 1,800 Days
• Field of Search: 704/200, 704/226, 704/227, 704/228, 704/233, 704/246, 704/248, 704/251, 704/253
• US Class Current: 704/233
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence

Abstract

A voice recognition method and apparatus in which an electrical signal corresponding to ambient noise is used to set a threshold value in accordance with the level of the ambient noise and a voice signal applied to a microphone is cut-out for processing if it exceeds the threshold value. The processing includes comparison of the voice signal cut-out with stored patterns of voice signals.

Claims

1. A voice recognizing apparatus, comprising:

a microphone for inputting a voice signal;

sound source means independent of any audio input signal received by said microphone for producing a variable level electrical signal representative of ambient noise to be applied to an audio sound source speakers to produce a variable level ambient audio noise that can be picked up by said microphone;

sampling means for sampling a frame of the voice signal input from said microphone that exceeds a threshold value and for sampling said electrical signs from said sound source means;

changing means receiving said sampled electrical signal for changing said threshold value in response to said sampled sound source means electrical signal; and

recognizing means for recognizing the voice signal above the threshold value sampled by said sampling means.

2. A voice recognizing apparatus in accordance with claim 1, further comprising invalidating means for invalidating said sampling means when the level of said electrical signal applied to said audio sound source exceeds a predetermined value.

3. A voice recognizing apparatus in accordance with claim 1, wherein said recognizing means includes a means for producing a determination standard of recognition, and standard changing means for changing said determination standard in accordance with said level of said electrical signal applied to said audio sound source.

4. A voice recognizing apparatus in accordance with claim 1 further comprising registering means for registering a reference pattern on the basis of the voice signal sampled by said sampling means.

5. A voice recognizing apparatus in accordance with claim 4, further comprising invalidating means for substantially invalidating said registering means when the level of said electrical signal applied to said audio sound source exceeds a predetermined value.

6. A voice recognizing apparatus in accordance with claim 1, wherein said changing means includes means for setting said threshold value in accordance with said level of the voice signal input to said microphone when the level of said electrical signal applied to said audio sound source is below a predetermined value.

7. A voice recognizing apparatus in accordance with claim 1, further comprising mode setting means for setting a first mode wherein the voice signal input to said microphone is to be processed or a second mode wherein the voice signal input to said microphone is not to be processed;

and means for setting a threshold value which is utilized in said first mode on the basis of the voice signal input to said microphone when said second mode is set.

8. A voice recognizing apparatus as in claim 1 wherein said voice signal and said electrical signal are sample at the same time.

9. A voice recognizing method comprising the steps of:

(a) receiving a voice signal input to a microphone;

(b) receiving from a sound source means independent of any signal received by said microphone a variable level electrical signal representative of ambient noise and applying said variable level electrical signal to an audio sound source to produce a variable level of ambient noise that can be received by said microphone;

(c) sampling the level of said electrical signal from said sound source means;

(d) sampling a voice signal input to said microphone which exceeds a threshold value;

(e) variably setting said threshold value for sampling said voice signal in accordance with said level of said sampled electrical signal from said sound source means; and

(f) recognizing the voice signal sampled in step (d).

10. A voice recognizing method in accordance with claim 9, further comprising a step of (g) registering a reference pattern on the basis of the voice signal as sampled.

11. A voice recognizing apparatus, comprising:

a microphone to which voice to be recognized is input;

first sampling means for sampling a feature parameter of a voice signal from said microphone for a plurality of frames, each of a predetermined time interval;

first converting means for converting said feature parameter sampled by said first sampling means into first feature parameter data;

first memory means for storing said first feature parameter data output from said first converting means for a plurality of frames;

first reading means for reading a series of said first feature parameter data exceeding a threshold value from said first memory means;

input means for receiving from a sound source means independent of any signal received by said microphone a variable level electrical signal representative of ambient noise to be supplied to an audio sound source that produces a corresponding variable level ambient audio noise to be received by said microphone;

detecting means for detecting from said sound source means electrical signal a level of ambient noise to be produced by the audio sound source;

threshold value setting means for variably setting said threshold value in accordance with an amplitude of the level of said sound source means electrical signal; and

recognition means for recognizing said voice on the basis of said series of said first feature parameter data.

12. A voice recognizing apparatus in accordance with claim 11, wherein said threshold value setting means includes means for setting said threshold value on the basis of power data of said voice signal inputted from said microphone when the level of said electrical signal is low.

13. A voice recognizing apparatus in accordance with claim 11, wherein said threshold value setting means includes weighted mean value calculation means for calculating a weighted mean value of said threshold value.

14. A voice recognizing apparatus in accordance with claim 12, further comprising second memory means for storing said threshold value set by said threshold value setting means for each frame;

wherein said weighted mean value calculation means calculates a weighted mean value of one or more threshold values of one or more prior frames read from said second memory means and a threshold value of a current frame.

15. A voice recognizing apparatus in accordance with claim 14, further comprising detecting means for detecting whether or not said level of said electrical signal increases;

wherein said weighted mean value calculation means sets a weight of the threshold value of said current frame to be larger than a weight of the threshold values of said prior frames when said detecting means detects that the level of said electrical signal increases.

16. A voice recognizing apparatus in accordance with claim 15, wherein said weighted mean value calculation means sets a weight of the threshold value of said current frame to be smaller than a weight of the threshold values of said prior frames when said detecting means does not detect that the level of said electrical signal increases.

17. A voice recognizing apparatus in accordance with claim 13, further comprising detecting means for detecting whether or not a level of said electrical signal increases;

wherein weighted mean value calculation means sets a weight of the threshold value of said current frame to be smaller than a weight of the threshold values of said prior frames when said detecting means does not detect that said levels of said electrical signal increases.

18. A voice recognizing apparatus in accordance with claim 11, further comprising invalidating means for invalidating said first reading means when the level of said electrical signal is more than a predetermined value.

19. A voice recognizing apparatus in accordance with claim 11, further comprising first and second address storing means for respectively storing a head frame and a tail frame of said first feature parameter data read from said first memory means by said first reading means.

20. A voice recognizing apparatus in accordance with claim 19, further comprising address determining means for determining a true head frame address and a true tail frame address on the basis of said head frame address and said tail frame address respectively stored in said first and second address storing means;

wherein said first reading means reads said first feature parameter data from said first memory means from said true head frame address to said true tail frame address.

21. A voice recognizing apparatus in accordance with claim 20, wherein said address determining means includes means for determining an address of said first memory means which is prior to said head frame address stored in said first address storing means and stores a minimum value as said true head frame address and an address of said first memory means which is after said head frame address stored in said first address storing means and stores a minimum value as said true tail frame address.

22. A voice recognizing apparatus in accordance with claim 11, further comprising:

second sampling means for sampling a feature parameter of said electrical signal representative of said ambient noise for each frame with said predetermined time interval;

second converting means for converting said feature parameter sampled by said second sampling means into second feature parameter data;

second memory means for storing said second feature parameter data outputted from said second converting means for a plurality of frames;

second reading means for reading said second feature parameter data from an address of said second memory means corresponding to that of said first memory means which is read by said first reading means; and

subtracting means for subtracting said second feature parameter data from said first feature parameter data.

23. A voice recognizing apparatus in accordance with claim 22, wherein said second reading means reads said second feature parameter data from an address of said second memory means equal to a frame which is delayed from a frame corresponding to an address of said first memory means which is read by said first reading means.

24. A voice recognizing apparatus in accordance with claim 22, wherein said subtracting means subtracts said second feature parameter data read from said second memory means which is multiplied by a predetermined constant from said first feature parameter data.

25. A voice recognizing apparatus in accordance with claim 22, wherein said recognition means recognizes the voice inputted from said microphone on the basis of a subtraction result by said subtracting means.

26. A voice recognizing apparatus in accordance with claim 25, further comprising first invalidating means for invalidating said recognition means when the level of said electrical signal is greater than a predetermined value.

27. A voice recognizing apparatus in accordance with claim 22, further comprising registration means for registering a feature parameter pattern of the voice inputted from said microphone on the basis of a subtraction result of said subtracting means.

28. A voice recognizing apparatus in accordance with claim 27, further comprising second invalidating means for invalidating said registration means when the level of said electrical signal is greater than a predetermined value.

29. A voice recognizing apparatus in accordance with claim 11, wherein the ambient noise is generated from a loudspeaker in response to said electrical signal produced from audio equipment.

30. A voice recognizing apparatus according to claim 29 wherein recognition means recognizes the voice inputted from said microphone through a comparison of said first feature parameter data read by said first reading means and a reference pattern.

31. A voice recognizing apparatus in accordance with claim 30, further comprising controlling means for controlling said audio equipment in response to a recognition result of said recognition means.

32. A voice recognizing apparatus in accordance with claim 11, wherein said threshold value setting means includes means for setting said threshold value in accordance with a level of a voice signal from said microphone when the level of said electrical signal is below a predetermined value.

33. A voice recognizing apparatus in accordance with claim 11, further comprising mode setting means for setting a first mode wherein the voice signal from said microphone is to be processed or a second mode wherein the voice signal from said microphone is not to be processed;

and means for setting a threshold value which is utilized in said first mode on the basis of the voice signal from said microphone when said second mode is set.

34. A voice recognizing apparatus in accordance with claim 22, wherein said ambient noise is generated from a loudspeaker in response to said electrical signal generated from audio equipment.