Speech signal processing apparatus for extracting a speech signal from a noisy speech signal

Claim 2

A signal processing apparatus comprising: a band division means for performing a band division process including a Fourier transformation for an inputted speech signal and for outputting spectrum signals of plural channels; a cepstrum analysis means for performing a cepstrum analysis process for the spectrum signals of plural channels outputted from said band division means and for outputting a cepstrum analysis result; a speech judgment means for detecting a speech signal interval of the inputted noisy speech signal in response to the cepstrum analysis result outputted from said cepstrum analysis means and for outputting the detected speech signal interval; and a speech extracting means for extracting a speech signal from the inputted noisy speech signal according to the detected speech signal interval outputted from said speech judgment means, and for outputting the extracted speech signal; wherein said speech judgment means comprises: a peak detection means for detecting a peak of a cepstrum in response to the cepstrum analysis result outputted from said cepstrum analysis means; an average value calculation means for calculating an average value of the cepstrum in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the calculated average value of the cepstrum; and a speech judgment circuit for detecting a speech signal interval in response to the detected peak of the cepstrum outputted from said peak detection means and the calculated average value of the cepstrum outputted from said average value calculation means; wherein said signal processing apparatus further comprises: a feature extraction means for extracting a speech feature from the extracted speech signal outputted from said speech extracting means, and for outputting the extracted speech feature; a storage means for initially storing standard speech features of plural speakers; and a feature comparison means for recognizing speech by comparing the extracted speech features outputted from said feature extraction means with the standard speech features stored in said storage means.

Claim 3

A signal processing apparatus comprising: a speech detection means for detecting a speech signal in response to an inputted noisy speech signal, and for outputting the detected speech signal; a noise prediction means for predicting speech noise in response to the inputted noisy speech signal according to the detected speech signal outputted from said speech detection means, and for outputting the predicted noise; a cancellation means for cancelling the predicted noise outputted from said noise prediction means from the inputted noisy speech signal, and for outputting the noise-canceled speech signal; and a speech extracting means for extracting a speech signal from the noise-canceled speech signal outputted from said cancellation means according to the detected speech signal outputted from said speech detection means.

Claim 4

A signal processing apparatus comprising: a speech detection means for detecting a speech signal in response to an inputted noisy speech signal, and for outputting the detected speech signal; a noise prediction means for predicting speech noise in response to the inputted noisy speech signal according to the detected speech signal outputted from said speech detection means, and for outputting the predicted noise; a cancellation means for cancelling the predicted noise outputted from said noise prediction means from the inputted noisy speech signal, and for outputting the noise-canceled speech signal; and a speech extracting means for extracting a speech signal from the noise-canceled speech signal outputted from said cancellation means according to the detected speech signal outputted from said speech detection means.

Claim 5

The signal processing apparatus as claimed in claim 2, further comprising: a band division means for performing a band division process including a Fourier transformation for an inputted speech signal and for outputting spectrum signals of plural channels to said cancellation means; and a band combining means for combining the noise-canceled spectrum signals of plural channels outputted from said cancellation means, inverse-Fourier-transforming the combined spectrum signal into a transformed speech signal, and for outputting the transformed speech signal to said speech cutting-out means; wherein said speech detection means comprises: a cepstrum analysis means for performing a cepstrum analysis process for the spectrum signals of plural channels outputted from said band division means, and for outputting the cepstrum analysis result; and a speech judgment means for detecting a speech signal in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the detected speech signal interval.

Claim 6

The signal processing apparatus as claimed in claim 2, further comprising: a feature extraction means for extracting a speech feature from the extracted speech signal outputted from said speech extracting means, and for outputting the extracted speech feature; a storage means for initially storing standard speech features of plural speakers; and a feature comparison means for recognizing speech by comparing the extracted speech features outputted from said feature extraction means with the standard speech features stored in said storage means.

Claim 7

The signal processing apparatus as claimed in claim 3, further comprising: a feature extraction means for extracting a speech feature from the extracted speech signal outputted from said speech extracting means, and for outputting the extracted speech feature; a storage means for initially storing standard speech features of plural speakers; and a feature comparison means for recognizing speech by comparing the extracted speech features outputted from said feature extraction means with the standard speech features stored in said storage means.

Claim 8

A signal processing apparatus comprising: an analog to digital convering means for converting an inputted noisy analog speech signal into a noisy digital speech signal, and for outputting the converted noisy digital speech signal; a Fourier transformation means for Fourier-transforming the converted noisy digital speech outputted from said analog to digital converting means into a transformed noisy digital spectrum signal, and for outputting the transformed noisy digital spectrum signal; a cepstrum analysis means for performing a cepstrum analysis process for the transformed noisy digital spectrum signal outputted from said Fourier transformation means, and for outputting the cepstrum analysis result; a speech judgment means for detecting a speech signal interval in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the detected speech signal interval; a noise interval judgement means for detecting a noise interval in response to the detected speech signal interval outputted from said speech judgment means; a muting means for attenuating the converted noisy digital speech signal only for the detected noise interval outputted form said noise interval judgment means according to the detected noise interval outputted from the noise interval judgement means, and for outputting the digital speech signal attenuated only for the detected noise interval; and a digital to analog converting means for converting the digital speech signal outputted from said muting means into an analog speech signal; wherein said speech judgment means comprises: a peak detection means for detecting a peak of a cepstrum in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputing the detected peak of the cepstrum; an average value calculation means for calculating an average value of the cepstrum in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the calculated average value; a vowel and consonant judgment means for detecting a vowel in response to the detected peak of the cepstrum outputted from said peak detection means and for detecting a consonant in response to the calculated average value outputted from said average value calculation means, and for outputting the detection result; and a speech judgment circuit for detecting a speech signal interval in response to the detection result outputted from said vowel and consonant judgment means, and for outputting the detected speech signal interval.

Claim 9

A signal processing apparatus comprising: an analog to digital convering means for converting an inputted noisy analog speech signal into a noisy digital speech signal, and for outputting the converted noisy digital speech signal; a Fourier transformation means for Fourier-transforming the converted noisy digital speech outputted from said analog to digital converting means into a transformed noisy digital spectrum signal, and for outputting the transformed noisy digital spectrum signal; a cepstrum analysis means for performing a cepstrum analysis process for the transformed noisy digital spectrum signal outputted from said Fourier transformation means, and for outputting the cepstrum analysis result; a speech judgment means for detecting a speech signal interval in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the detected speech signal interval; a noise interval judgement means for detecting a noise interval in response to the detected speech signal interval outputted from said speech judgment means; a muting means for attenuating the converted noisy digital speech signal only for the detected noise interval outputted form said noise interval judgment means according to the detected noise interval outputted from the noise interval judgement means, and for outputting the digital speech signal attenuated only for the detected noise interval; and a digital to analog converting means for converting the digital speech signal outputted from said muting means into an analog speech signal; wherein said speech judgment means comprises: a peak detection means for detecting a peak of a cepstrum in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputing the detected peak of the cepstrum; an average value calculation means for calculating an average value of the cepstrum in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the calculated average value; a vowel and consonant judgment means for detecting a vowel in response to the detected peak of the cepstrum outputted from said peak detection means and for detecting a consonant in response to the calculated average value outputted from said average value calculation means, and for outputting the detection result; and a speech judgment circuit for detecting a speech signal interval in response to the detection result outputted from said vowel and consonant judgment means, and for outputting the detected speech signal interval.

Claim 10

A signal processing apparatus comprising: a band division means for performing a band division process including a Fourier transformation for an inputted speech signal and for outputting spectrum signals of plural channels; a cepstrum analysis means for performing a cepstrum analysis process for the spectrum signals of plural channels outputted from said band division means and for outputting a cepstrum analysis result; a speech judgment means for detecting a speech signal interval in response to the cepstrum analysis result outputted from said cepstrum analysis mean, and for outputting the detected speech signal interval; a noise interval judgment means for detecting a noise interval in response to the detected speech signal interval outputted from said speech judgment means; and a muting means for attenuating the inputted noisy speech signal only for the detected noise interval outputted from said noise interval judgment means according to the detected noise interval outputted from said noise interval judgement means, and for outputting the speech signal attenuated only for the detected noise interval; wherein said speech judgment means comprises: a peak detection means for detecting a peak of a cepstrum in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the detected peak of the cepstrum; an average value calculation means for calculating an average value of a cepstrum in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the calculated average value; a vowel and consonant judgment means for detecting a vowel in response to the detected peak of the cepstrum outputted from said peak detection means and detecting a consonant in response to the calculated average value outputted from said average value calculation means, and for outputting the detection result; and a speech judgment circuit for detecting a speech signal interval in response to the detection result outputted from said vowel and consonant judgment means, and for outputting the detected speech signal interval.

Claim 11

A signal processing apparatus comprising: a band division means for performing a band division process including a Fourier transformation for an inputted speech signal and for outputting spectrum signals of plural channels; a cepstrum analysis means for performing a cepstrum analysis process for the spectrum signals of plural channels outputted from said band division means and for outputting a cepstrum analysis result; a speech judgment means for detecting a speech signal interval in response to the cepstrum analysis result outputted from said cepstrum analysis mean, and for outputting the detected speech signal interval; a noise interval judgment means for detecting a noise interval in response to the detected speech signal interval outputted from said speech judgment means; and a muting means for attenuating the inputted noisy speech signal only for the detected noise interval outputted from said noise interval judgment means according to the detected noise interval outputted from said noise interval judgement means, and for outputting the speech signal attenuated only for the detected noise interval; wherein said speech judgment means comprises: a peak detection means for detecting a peak of a cepstrum in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the detected peak of the cepstrum; an average value calculation means for calculating an average value of a cepstrum in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the calculated average value; a vowel and consonant judgment means for detecting a vowel in response to the detected peak of the cepstrum outputted from said peak detection means and detecting a consonant in response to the calculated average value outputted from said average value calculation means, and for outputting the detection result; and a speech judgment circuit for detecting a speech signal interval in response to the detection result outputted from said vowel and consonant judgment means, and for outputting the detected speech signal interval.

Claim 12

A signal processing apparatus comprising: a storage means for initially storing speech features of plural speakers; a speech detection means for detecting a speech signal in response to an inputted noisy speech signal, and for outputting the detected speech signal interval; a maximum likelihood estimation means for detecting a kind of speech by comparing eh detected feature of the speech signal outputted from said speech detection means with the speech features of plural speakers stored in said storage means, and for outputting the detected kind of speech; a noise interval judgment means for detecting a noise interval in response to the detected kind of speech outputted from said maximum likelihood estimation means and the detected speech signal interval outputted from said speech detection means, and for outputting the detection noise interval; and a muting means for attenuation the inputted noisy speech signal only for the detected noise interval outputted from said noise interval judgment means according to the detected noise interval outputted rom said noise interval judgment means, and for outputting the speech signal attenuated only for the detected noise interval.

Claim 13

A signal processing apparatus comprising: a storage means for initially storing speech features of plural speakers; a speech detection means for detecting a speech signal in response to an inputted noisy speech signal, and for outputting the detected speech signal interval; a maximum likelihood estimation means for detecting a kind of speech by comparing eh detected feature of the speech signal outputted from said speech detection means with the speech features of plural speakers stored in said storage means, and for outputting the detected kind of speech; a noise interval judgment means for detecting a noise interval in response to the detected kind of speech outputted from said maximum likelihood estimation means and the detected speech signal interval outputted from said speech detection means, and for outputting the detection noise interval; and a muting means for attenuation the inputted noisy speech signal only for the detected noise interval outputted from said noise interval judgment means according to the detected noise interval outputted rom said noise interval judgment means, and for outputting the speech signal attenuated only for the detected noise interval.

Claim 14

The signal processing apparatus as claimed in claim 8, wherein said speech detection means comprises: a band division means for performing a band division process including a Fourier transformation for an inputted speech signal and for outputting spectrum signals of plural channels; a cepstrum analysis means for performing a cepstrum analysis process for the spectrum signals of plural channels outputted from said band division means and for outputting a cepstrum analysis result; a speech judgment means for detecting a speech signal in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the detected speech signal interval.

Claim 15

The signal processing apparatus as claimed in claim 9, wherein said speech judgment means comprises: a peak detection means for detecting a peak of a cepstrum in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the detected peak of the cepstrum to said maximum likelihood estimation means; an average value calculation means for calculating an average value of the cepstrum in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the calculated average value to said maximum likelihood estimation means; a vowel and consonant judgment means for detecting a vowel in response to the detected peak of the cepstrum outputted from said peak detection means and for detecting a consonant in response to the calculated average value outputted from said average value calculation means, and for outputting the detection result; and a speech judgment circuit for detecting a speech signal in response to the detection result outputted from said vowel and consonant judgment means, and for outputting the detected speech signal interval; wherein the speech features of plural speakers stored in said storage means and the speech features compared by said maximum likelihood estimation means are peaks of the cepstrum and average values of the cepstrums.

Claim 16

A signal processing apparatus comprising: a speech detection means for detecting a speech signal interval in response to an inputted noisy speech signal and for outputting the detected speech signal interval; a noise interval judgement means for detecting a noise interval in response to the detected speech signal interval outputted from said speech detection means, and for outputting the detected noise interval; a noise prediction means for predicting a noise of the inputted noisy speech signal interval in response to the detected noise interval outputted from said noise interval judgment means, and for outputting the predicted noise; a cancellation means for canceling the predicted noise outputted from said noise prediction means in the inputted noisy speech signal and for outputting a nose-canceled speech signal; and a muting means for attenuating the noise-canceled speech signal outputted form said cancellation means, only for the detected noise interval outputted from said noise interval judgment means with a predetermined attenuation value according to the detecting noise interval outputted from said noise interval judgment means, and for outputting the speech signal attenuated only for the detected noise interval.

Claim 17

A signal processing apparatus comprising: a speech detection means for detecting a speech signal interval in response to an inputted noisy speech signal and for outputting the detected speech signal interval; a noise interval judgement means for detecting a noise interval in response to the detected speech signal interval outputted from said speech detection means, and for outputting the detected noise interval; a noise prediction means for predicting a noise of the inputted noisy speech signal interval in response to the detected noise interval outputted from said noise interval judgment means, and for outputting the predicted noise; a cancellation means for canceling the predicted noise outputted from said noise prediction means in the inputted noisy speech signal and for outputting a nose-canceled speech signal; and a muting means for attenuating the noise-canceled speech signal outputted form said cancellation means, only for the detected noise interval outputted from said noise interval judgment means with a predetermined attenuation value according to the detecting noise interval outputted from said noise interval judgment means, and for outputting the speech signal attenuated only for the detected noise interval.

Claim 18

The signal processing apparatus as claimed in claim 11, further comprising a noise power calculation means for calculating a power of a noise in response to the inputted noisy speech signal and the detected speech signal interval outputted from said speech detection means and for outputting the calculated power of the noise to said muting means; wherein the attenuation value of said muting means is altered depending on the calculated power of the noise outputted from said noise power calculation means.

Claim 19

A signal processing apparatus comprising: a band division means for performing a band division process including a Fourier transformation for an inputted speech signal and for outputting spectrum signals of plural channels; a cepstrum analysis means for performing a cepstrum analysis process for the spectrum signals of plural channels outputted from said band division means, and for outputting the cepstrum analysis result; a speech judgment means for detecting a speech signal interval in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the detected speech signal interval; a noise interval judgement means for detecting a noise interval in response to the detected speech signal interval outputted from said speech judgment means, and for outputting the detected noise interval; a noise prediction means for predicting noise of the spectrum signals of plural channels outputted from said band division means in response to the detected noise interval outputted from said noise interval judgement means, and for outputting the predetected noise of plural channels; a cancellation means for canceling the predicted noise of plural channels outputted form said noise prediction means in the spectrum signals of plural channels outputted from said band division means, and for outputting noise-canceled spectrum signals of plural channels; a band combining means for combining the noise-canceled spectrum signals of plural channels, inverse-Fourier-transforming the combined spectrum signal into a transformed speech signal, and for outputting the transformed speech signal; and a muting means for attenuating the transformed speech signal outputted from said band combining means, only for the detected noise interval outputted from said noise interval judgment means with a predetermined attenuation value according to the detected noise interval outputted from said noise interval judgment means, and for outputting the speech signal attenuated only for the detected noise interval.

Claim 20

A signal processing apparatus comprising: a band division means for performing a band division process including a Fourier transformation for an inputted speech signal and for outputting spectrum signals of plural channels; a cepstrum analysis means for performing a cepstrum analysis process for the spectrum signals of plural channels outputted from said band division means, and for outputting the cepstrum analysis result; a speech judgment means for detecting a speech signal interval in response to the cepstrum analysis result outputted from said cepstrum analysis means, and for outputting the detected speech signal interval; a noise interval judgement means for detecting a noise interval in response to the detected speech signal interval outputted from said speech judgment means, and for outputting the detected noise interval; a noise prediction means for predicting noise of the spectrum signals of plural channels outputted from said band division means in response to the detected noise interval outputted from said noise interval judgement means, and for outputting the predetected noise of plural channels; a cancellation means for canceling the predicted noise of plural channels outputted form said noise prediction means in the spectrum signals of plural channels outputted from said band division means, and for outputting noise-canceled spectrum signals of plural channels; a band combining means for combining the noise-canceled spectrum signals of plural channels, inverse-Fourier-transforming the combined spectrum signal into a transformed speech signal, and for outputting the transformed speech signal; and a muting means for attenuating the transformed speech signal outputted from said band combining means, only for the detected noise interval outputted from said noise interval judgment means with a predetermined attenuation value according to the detected noise interval outputted from said noise interval judgment means, and for outputting the speech signal attenuated only for the detected noise interval.