Apparatus and method for speech signal analysis

• US 5,680,506 A
• Filed: 12/29/1994
• Issued: 10/21/1997
• Est. Priority Date: 12/29/1994
• Status: Expired due to Term

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Lucent Technologies, Inc. (Nokia Corporation)
• Inventors: Pai, Suhas A., Soong, Frank Kao-Ping, Kroon, Peter
• Primary Examiner(s): Zele, Krista M.
• Assistant Examiner(s): Wolinsky, Scott
• Application Number:
  US08/368,059
• Time in Patent Office: 1,027 Days
• Field of Search: 379/58, 379/59, 379/63, 379/88, 379/89, 395/2.16, 395/2.28, 395/2.38, 395/2.53, 395/2.62, 395/2.55, 395/2.65, 395/2.3, 395/2.2, 395/2.26, 395/2.63, 395/2.12, 381/36, 381/37, 381/41, 381/43, 381/49, 370/110.3
• US Class Current: 704/203
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 19/16 Vocoder architecture
• CPC Class Codes: G10L 25/12 the extracted parameters being prediction coefficients
• CPC Class Codes: G10L 25/24 the extracted parameters being the cepstrum

Abstract

The present invention provides a novel method of analyzing speech signals in order to reduce the computational power required to perform both speech compression and voice recognition operations. Digital speech signals are provided to a speech analyzer which generates a linear predictive coded (LPC) speech analysis signal that is compatible for use in both the voice recognition circuit and the speech compression circuit. The speech analysis signal is then provided to the compression circuit, which further processes the signal into a form used by an encoder and then the encoder encodes the processed signal. The same speech analysis signal is also provided to a voice recognition circuit, which further processes the signal into a form used by a recognizer and then the recognizer performs recognition on the processed signal.

Claims

1. A method for providing a speech analysis signal for use in a speech compression circuit including an encoder and a voice recognition circuit including a recognizer, the method comprising the steps of:

a) receiving a digital speech signal;

b) analyzing the digital speech signal and producing a speech analysis signal therefrom, said speech analysis signal compatible for use within the speech compression circuit and the voice recognition circuit;

c) providing said speech analysis signal to the speech compression circuit for subsequent transformation into a form compatible with the encoder; and

d) providing said speech analysis signal to the voice recognition circuit for subsequent transformation into a form compatible with the recognizer.

2. The method of claim 1 wherein step b) includes the steps of:

defining a plurality of windows, each window comprising a temporal portion of the digital speech signal, such that the plurality of windows comprise the digital speech signal; and

generating a speech analysis signal by generating a vector for each of the plurality of windows.

3. The method of claim 2 wherein the plurality of windows comprises a plurality of non-overlapping windows.

4. The method of claim 2 wherein the step of generating a vector further comprises generating a set of autocorrelation coefficients for each defined window.

5. The method of claim 2 wherein the step of defining a plurality of windows comprises defining a plurality of non-overlapping windows, each window having a duration of substantially 20 msecs.

6. The method of claim 1 wherein the step of producing a speech analysis signal includes producing a linear predictive coded (LPC) speech analysis signal.

7. A method of processing speech signals comprising the steps of:

a) receiving an analog input speech signal;

b) converting the analog input speech signal to a digital speech signal;

c) analyzing the digital speech signal and producing a speech analysis signal therefrom, said speech analysis signal comprising a plurality of analysis vectors;

d) performing speech compression transformations on the plurality of analysis vectors to produce a plurality of compression transformation vectors;

e) employing an encoder to produce a compressed speech signal using the compression transformation vectors;

f) performing voice recognition transformations on the plurality of analysis vectors to produce a plurality of observation vectors; and

g) performing voice recognition using the observation vectors.

8. The method of claim 7 wherein step c) further comprises:

defining a plurality of windows, each window comprising a portion of the digital speech signal, such that the plurality of windows comprise the digital speech signal; and

generating a speech analysis signal by generating an analysis vector for each defined window.

9. The method of claim 8 wherein the plurality of windows comprises a plurality of non-overlapping windows.

10. The method of claim 8 wherein the step of generating an analysis vector further comprises generating a set of autocorrelation coefficients for each defined window.

11. The method of claim 10 wherein step d) further comprises generating reflection coefficients from the autocorrelation coefficients for each defined window.

12. The method of claim 10 wherein step f) further comprises:

generating a plurality linear predictive coding coefficients from the plurality of autocorrelation coefficients; and

generating a plurality of cepstral coefficients from the plurality of linear predictive coding coefficients.

13. The method of claim 7 further comprising the step of filtering the digital speech signal prior to performing step c).

14. The method of claim 13 wherein the step of filtering the digital speech signal further comprises effecting high pass filtering and preemphasis filtering on the digital speech signal.

15. The method of claim 7 wherein step g) further comprises performing hidden Markov model pattern matching between a stored plurality of model vectors and the plurality of observation vectors.

16. A method of processing a compressed digital speech signal, said compressed digital signal being generated by a remotely-located user, the method comprising the steps of:

a) receiving the compressed digital speech signal;

b) decoding the compressed digital speech signal to produce a compression transform signal;

c) inversely transforming the compression transform signal to produce a speech analysis signal comprising a plurality of speech analysis vectors;

d) performing voice recognition transformations on the plurality of analysis vectors to produce a plurality of observation vectors; and

e) performing voice recognition on said observation vectors to produce a recognition output signal therefrom.

17. The method of claim 16 wherein step b) further comprises the steps of dequantizing the compressed digital speech signal and interpolating the dequantized compressed digital speech signal.

18. The method of claim 16 wherein step c) further comprises:

generating a plurality of linear predictive coding coefficients using the plurality of analysis vectors; and

generating a plurality of cepstral coefficients from the plurality of linear predictive coding coefficients.

19. The method of claim 16 wherein step d) further comprises performing hidden Markov model pattern matching between a plurality of codebook vectors and the plurality of observation vectors.

20. The method of claim 16 wherein each of said analysis vectors comprises a plurality of autocorrelation coefficients representative of a spectral analysis of a portion of the digital speech signal.

21. The method of claim 16 further comprising the step of using the recognition output signal to cause a selected operation to be performed.

22. The method of claim 16 further comprising the step of using the recognition output signal to identify the remotely-located user.

23. An apparatus for processing digital speech signals comprising:

a) a speech analyzer having an input for receiving a digital speech signal,

said speech analyzer operating to define a plurality of windows, each window corresponding to a portion of the digital speech signal, means operably connected to the input;

and to generate an analysis vector for each window on an output, said analysis vector representative of the spectral characteristics of the portion of the speech signal corresponding to the window;

b) a speech recognition circuit connected to the output of the speech analyzer for processing said analysis vectors; and

c) a speech compression circuit connected to the output of the speech analyzer for processing said analysis vectors.

24. The apparatus of claim 23 wherein the plurality of windows comprises a plurality of non-overlapping windows.

25. The apparatus of claim 23 wherein the speech analyzer further comprises a programmed digital signal processor.

26. An apparatus for processing speech signals comprising:

a) a speech analyzer having an input and an output, the speech analyzer operable to receive a speech signal at the input and produce a speech analysis signal at the output;

b) a compression transform unit connected to the speech analyzer output and operable to transform the speech analysis signal into a compression transform signal for use in an encoder circuit;

c) an encoder circuit connected to the compression transform circuit to receive the compression transform signal therefrom, and operable to encode the compression transform signal;

d) a recognition transform unit connected to the speech analyzer output and operable to transform the speech analysis signal into an observation sequence for use in a recognizer; and

e) the recognizer connected to the recognition transform unit to receive the observation sequence therefrom.

27. The apparatus of claim 26 wherein the speech analyzer

defines a plurality of windows, each window comprising a portion of the speech signal, such that the plurality of windows comprise the speech signal;

generates an analysis vector for each defined window; and

generates a speech analysis signal comprising the plurality of generated analysis vectors.

28. The apparatus of claim 27 wherein the plurality of windows comprise a plurality of non-overlapping windows.

29. The apparatus of claim 26 wherein each of the analysis vectors includes a plurality of linear predictive coding (LPC) coefficients and the compression transform unit is further operable to transform LPC coefficients into reflection coefficients.

30. The apparatus of claim 26 wherein each of the analysis vectors includes a plurality of linear predictive coding (LPC) coefficients and the recognition transform unit is further operable to transform LPC coefficients into cepstral coefficients.

31. The apparatus of claim 26 wherein each of the analysis vectors includes a plurality of linear predictive coding (LPC) coefficients and the recognition transform unit is operable to:

transform LPC coefficients into cepstral coefficients; and

transform LPC coefficients into delta cepstrum coefficients.

32. The apparatus of claim 26 wherein the recognizer comprises a hidden Markov model recognizer.

33. The apparatus of claim 26 wherein the apparatus for processing speech signals comprises a digital cellular telephone.

34. An apparatus for processing analog speech signals in a digital cellular telephone, said apparatus comprising:

a) an analog-to-digital converter;

b) a speech analyzer having an input and an output, the input connected to the analog-to-digital converter, the speech analyzer operable to receive a speech signal at the input and produce a speech analysis signal at the output, the speech analysis signal comprising a plurality of linear predictive coding (LPC) vectors;

c) a compression transform unit connected to the speech analyzer output and operable to transform the speech analysis signal into a compression transform signal for use in an encoder circuit;

d) an encoder circuit connected to the compression transform circuit to receive the compression transform signal therefrom, and operable to encode the compression transform signal;

e) a recognition transform unit connected to the speech analyzer output and operable to transform the speech analysis signal into an observation sequence for use in a recognizer;

f) a recognizer connected to the recognition transform unit to receive the observation sequence therefrom; and

g) a transmitter connected to the encoder circuit.

35. An apparatus for transmitting compressed digital speech signals comprising:

a) a speech analyzer having an input and an output, the speech analyzer operable to receive a speech signal at the input and produce a speech analysis signal at the output, the speech analysis signal comprising a plurality of linear predictive coding (LPC) vectors, said speech analysis signal compatible for use in a remotely-located voice recognizer;

b) a compression transform unit connected to the speech analyzer output and operable to transform the speech analysis signal into a format suitable for use in an encoder circuit;

c) an encoder circuit connected to the compression transform circuit to receive transformed speech analysis signals therefrom; and

d) a transmitter connected to the encoder circuit, the transmitter operable to transmit an encoded signal to a receiver circuit including the remotely-located voice recognizer.

36. The apparatus of claim 35 wherein the speech analyzer is operable to:

define a plurality of windows, each window comprising a portion of the speech signal, such that the plurality of windows comprise the speech signal;

and

generate a speech analysis signal comprising a plurality of LPC vectors with each of said plurality of LPC vectors generated for a corresponding one of the plurality of windows.

37. The apparatus of claim 36 wherein the plurality of windows comprise a plurality of non-overlapping windows.

38. An apparatus for receiving and processing compressed digital speech signals comprising:

a) a signal receiver operable to receive compressed digital speech signals from a remotely located transmitter and to produce at an output received compressed digital speech signals;

b) a compression inverse transform unit connected to the output of the signal receiver and operable to transform the received compressed digital speech signals into linear predictive coding (LPC) vector signals;

c) a recognition transform unit connected to the compression inverse transform unit and operable to produce transformed LPC vector signals in a format compatible for use in a recognizer;

d) the recognizer connected to the output of the recognition transform unit to receive the transformed speech analysis signals therefrom.

39. An apparatus for providing speech signal processing in a digital cellular telephone comprising:

a) a microphone;

b) an analog-to-digital converter connected to the microphone;

c) a digital signal processor operably connected to receive a digital speech signal from the analog-to-digital converter, the digital signal processor operable to perform speech compression on the digital speech signal to produce an encoded signal, and further operable to perform voice recognition on the digital speech signal to produce a recognized signal; and

d) a transmitter operably connected to receive the encoded signal from the digital signal processor.

40. The apparatus of claim 39 further comprising a first memory connected to the digital speech processor for storing a plurality of speech models and wherein the digital signal processor uses the speech models to perform voice recognition.

41. The apparatus of claim 39 wherein the digital cellular telephone is operable to receive a voice command and the recognized signal comprises a signal representative of a lexicon of the voice command.