Vector quantizing apparatus and speech analysis-synthesis system using the apparatus

Claim 2

A speech analysis-synthesis system for synthesizing by using a small amount of information, the system comprising: a separation means for separating a speech signal into short-time speech data; a pitch extraction means for applying pitch analysis to said short-time speech data to output a pitch; a spectral analysis means for applying spectral analysis to said short-time speech data to output a row of logarithmic power spectral data; a spectral envelope generation means for generating a spectral envelope from said row of logarithmic power spectral data; a spectral envelope vector conversion means for normalizing logarithmic power values of said spectral envelope to extract a predetermined number of logarithmic power values as spectral envelope vectors; a vector quantization means for quantizing said spectral envelope vectors by comparison with a plurality of vectors prepared in advance; a spectral envelope reconstruction means for reconstructing said spectral envelope by interpolation of said quantized spectral envelope vectors; an impulse response conversion means for converting said reconstructed spectral envelope into an impulse response waveform; and a waveform synthesis means for synthesizing a speech waveform by superposition at intervals of said pitch while compensating an amplitude of said impulse response waveform.

Claim 3

A speech analysis-synthesis system according to claim 1, wherein said spectral envelope vector conversion means includes means for extracting a predetermined number of logarithmic power values at intervals so that a low-frequency portion becomes more dense while a high-frequency portion becomes less dense, from spectral envelope samples after normalization of logarithmic power values.

Claim 4

A speech analysis-synthesis system according to claim 1, wherein said spectral envelope reconstruction means includes means for performing parabolic interpolation on each adjacent three samples of said spectral envelope vectors quantized by said vector quantization means, performing interpolation into space between two samples by adding weighted adjacent two parabolas, and reconstructing said logarithmic power values by an operation reverse to normalization of said logarithmic power values to thereby reconstruct said spectral envelope.

Claim 5

A speech analysis-synthesis system for synthesizing by using a small amount of information, the system comprising: a separation means for separating a speech signal into short-time speech data; a pitch extraction means for performing pitch analysis on said short-time speech data to thereby output a pitch; a spectral analysis means for performing spectral analysis on said short-time speech data to thereby output a row of logarithmic power spectral data; a spectral envelope generation means for generating a spectral envelope from said row of logarithmic power spectral data; an impulse response conversion means for converting said spectral envelope into an impulse response waveform; an impulse response vector conversion means for normalizing an amplitude of said impulse response waveform to extract a predetermined number of amplitude data as impulse response vectors; a vector quantization means for quantizing said impulse response vectors by comparison with a plurality of vectors prepared in advance; an impulse response reconstruction means for reconstructing said impulse response waveform by interpolation or extrapolation of said quantized impulse response vectors; and a waveform synthesis means for synthesizing a voice waveform by superposition at intervals of said pitch while compensating the amplitude of said reconstructed impulse response waveform.

Claim 6

A speech analysis-synthesis system for synthesizing by using a small amount of information, the system comprising: a separation means for separating a speech signal into short-time speech data; a pitch extraction means for performing pitch analysis on said short-time speech data to thereby output a pitch; a spectral analysis means for performing spectral analysis on said short-time speech data to thereby output a row of logarithmic power spectral data; a spectral envelope generation means for generating a spectral envelope from said row of logarithmic power spectral data; an impulse response conversion means for converting said spectral envelope into an impulse response waveform; an impulse response vector conversion means for normalizing an amplitude of said impulse response waveform to extract a predetermined number of amplitude data as impulse response vectors; a vector quantization means for quantizing said impulse response vectors by comparison with a plurality of vectors prepared in advance; an impulse response reconstruction means for reconstructing said impulse response waveform by interpolation or extrapolation of said quantized impulse response vectors; and a waveform synthesis means for synthesizing a voice waveform by superposition at intervals of said pitch while compensating the amplitude of said reconstructed impulse response waveform.

Claim 7

A speech analysis-synthesis system according to claim 4, wherein said impulse response vector conversion means includes means for extracting the predetermined number of amplitude data from samples of said impulse response waveform after normalization of the amplitude thereof, according to an extracting manner determined on the basis of the value of said pitch in advance.

Claim 8

A speech analysis-synthesis system according to claim 4, wherein said impulse response reconstruction means includes means for performing linear predictive analysis on said impulse response vectors quantized by said vector quantization means while regarding said impulse response vectors as waveforms, extrapolating said quantized impulse response vectors by using a linear predictive coefficient, and reconstructing said amplitude by an operation reverse to normalization of said amplitude to thereby reconstruct said impulse response waveform.

Claim 9

A speech analysis-synthesis system according to claim 4, wherein said spectral envelope generation means includes means for performing parabolic interpolation to each adjacent three data of said logarithmic power spectral data row, and performing interpolation into space between two data by adding weighted adjacent two parabolas to thereby generate said spectral envelope.

Claim 10

A speech analysis-synthesis system according to claim 4, wherein said vector quantization means includes means for using fuzzy vector quantization or complementary vector quantization in which spectral envelope vectors or said impulse response vectors are quantized by a linear combination of at least two vectors in a plurality of vectors prepared in advance.

Claim 11

A speech coding/decoding system using a speech analysis-synthesis system according to claim 8, wherein said vector quantization means is separated into a vector quantization coding means for receiving said spectral envelope vectors or said impulse response vectors and for outputting indexes and attendant information corresponding to said spectral envelope vectors or said impulse response vectors, and a vector quantization decoding means for decoding the quantized values of said spectral envelope vectors or said impulse response vectors by using said plurality of vectors prepared in advance on the basis of said indexes and attendant information, and further wherein means from said separation means to said vector quantization coding means are provided in a transmitter side and means from said vector quantization decoding means to said speech waveform synthesis means are provided in a receiver side.