Speech recognition system and standard pattern preparation system as well as speech recognition method and standard pattern preparation method

• US 6,741,962 B2
• Filed: 03/07/2002
• Issued: 05/25/2004
• Est. Priority Date: 03/08/2001
• Status: Expired due to Fees

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Iso, Kenichi
• Primary Examiner(s): Dorvil, Richemond
• Assistant Examiner(s): Han, Qi
• Application Number:
  US10/093,110
• Publication Number:
  US 20020128835A1
• Time in Patent Office: 810 Days
• Field of Search: 704/220, 704/238, 704/207, 704/247, 704/268, 704/203, 704/253, 704/223
• US Class Current: 704/247
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 21/038 using band spreading techniques

Abstract

A speech recognition system for recognizing an input voice of a narrow frequency band. The speech recognition system includes: a frequency band converting unit for converting the input voice of the narrow frequency band into a pseudo voice of a wide frequency band which covers an entirety of the narrow frequency band and which is wider than the narrow frequency band.

Claims

1. A speech recognition system for recognizing an input voice of a narrow frequency band, said speech recognition system including:

a power spectrum calculating unit for calculating power spectrums of said input voice of said narrow frequency band;

a frequency band converting unit for converting said input voice of said narrow frequency band into a pseudo voice of a wide frequency band which covers an entirety of said narrow frequency band and which is wider than said narrow frequency band, said frequency band converting unit comprising;

i. a eigen-vector storing unit for storing a plurality of eigen vectors of power spectrums of said wide frequency band pseudo voice;

ii. an expansion coefficient calculating unit for calculating expansion coefficients that said power spectrums calculated by said power spectrum calculating unit are expanded by a linear combination of said plurality of eigen vectors;

iii. a frequency band expansion unit for calculating additional power spectrums in a lack frequency band by use of said expansion coefficients calculated by said expansion coefficient calculating unit, where said lack frequency band is covered by said wide frequency band but not covered by said narrow frequency band, and said frequency band expansion unit combining said additional power spectrums in said lack frequency band into said power spectrum of said narrow frequency band calculated by said power spectrum calculating unit to prepare pseudo power spectrums of said pseudo voice of said wide frequency band; and

iv. a melcepstrum calculating unit for calculating a melcepstrum characteristic quantity based on said pseudo power spectrum prepared by said frequency band expansion unit;

a pattern reference unit for receiving an output from said frequency band converting unit and comparing said output with a standard pattern for carrying out the speech recognition;

a standard pattern preparation unit for receiving an output from said frequency band converting unit and preparing the standard pattern based on said output for carrying out the speech recognition; and

a standard pattern storing unit for storing the standard pattern prepared by the standard pattern preparation unit for allowing said pattern reference unit to compare said output with said standard pattern.

2. A standard pattern preparation system for preparing a standard pattern to recognize an input voice of a narrow frequency band, said standard pattern preparation system including:

a power spectrum calculating unit for calculating power spectrums of said input voice of said narrow frequency band;

a frequency band converting unit for converting said input voice of said narrow frequency band into a pseudo voice of a wide frequency band which covers an entirety of said narrow frequency band and which is wider than said narrow frequency band, said frequency band converting unit comprising;

i. an eigen vector storing unit for storing a plurality of eigen vectors of power spectrums of said wide frequency band voice;

ii. an expansion coefficient calculating unit for calculating expansion coefficients that said power spectrums calculated by said power spectrum calculating unit are expanded by a linear combination of said plurality of eigen vectors; and

iii. a frequency band expansion unit for calculating additional power spectrums in a lack frequency band by use of said expansion coefficients calculated by said expansion coefficient calculating unit, where said lack frequency band is covered by said wide frequency band but not covered by said narrow frequency band, and said frequency band expansion unit combining said additional power spectrums in said lack frequency band into said power spectrum of said narrow frequency band calculated by said power spectrum calculating unit to prepare pseudo power spectrums of said pseudo voice of said wide frequency band; and

iv. a melcepstrum calculating unit for calculating a melcepstrum characteristic quantity based on said pseudo power spectrum prepared by said frequency band expansion unit;

a pattern reference unit for receiving an output from said frequency band converting unit and comparing said output with a standard pattern for carrying out the speech recognition;

a standard pattern preparation unit for receiving an output from said frequency band converting unit and preparing the standard pattern based on said output for carrying out the speech recognition; and

a standard pattern storing unit for storing the standard pattern prepared by the standard pattern preparation unit for allowing said pattern reference unit to compare said output with said standard pattern.

3. A speech recognition method for recognizing an input voice of a narrow frequency band, said speech recognition method including the steps of:

a. calculating power spectrums of said input voice of said narrow frequency band;

b. converting said input voice of said narrow frequency band into a pseudo voice of a wide frequency band which covers an entirety of said narrow frequency band and which is wider than said narrow frequency band;

c. calculating expansion coefficients that said power spectrums are expanded by a linear combination of a plurality of eigen vectors of said power spectrums of said wide frequency band voice;

d. calculating additional power spectrums in a lack frequency band by use of said expansion coefficients, where said lack frequency band is covered by said wide frequency band but not covered by said narrow frequency band, so as to combine said additional power spectrums in said lack frequency band into said power spectrum of said narrow frequency band thereby to prepare pseudo power spectrums of said pseudo voice of said wide frequency band;

e. calculating a melcepstrum characteristic quantity based on said pseudo power spectrum; and

f. comparing said melcepstrum characteristic quantity with a standard pattern for carrying out the speech recognition.

4. A standard pattern preparation method for preparing a standard pattern to recognize an input voice of a narrow frequency band, said standard pattern preparation method including the steps of:

a. calculating power spectrums of said input voice of said narrow frequency band;

b. converting said input voice of said narrow frequency band into a pseudo voice of a wide frequency band which covers an entirety of said narrow frequency band and which is wider than said narrow frequency band; and

c. calculating expansion coefficients that said power spectrums are expanded by a linear combination of a plurality of eigen vectors of said power spectrums of said wide frequency band voice;

d. calculating additional power spectrums in a lack frequency band by use of said expansion coefficients, where said lack frequency band is covered by said wide frequency band but not covered by said narrow frequency bands and combining said additional power spectrums in said lack frequency band into said power spectrum of said narrow frequency band to prepare pseudo power spectrums of said pseudo voice of said wide frequency band;

e. receiving said pseudo voice and preparing a standard pattern based on said output for carrying out speech recognition;

f. calculating a melcepstrum characteristic quantity based on said pseudo power spectrum; and

g. comparing said melcepstrum characteristic quantity with a standard pattern for carrying out the speech recognition.