Speech enhancement
View Patent ↗A speech signal processing system is described for use with automatic speech recognition and hands free speech communication. A signal pre-processor module transforms an input microphone signal into corresponding speech component signals. A noise suppression module applies noise reduction to the speech component signals to generate noise reduced speech component signals. A speech reconstruction module produces corresponding synthesized speech component signals for distorted speech component signals. A signal combination block adaptively combines the noise reduced speech component signals and the synthesized speech component signals based on signal to noise conditions to generate enhanced speech component signals for automatic speech recognition and hands free speech communication.
1. A computer-implemented method employing at least one hardware implemented computer processor for speech signal processing for automatic speech recognition and hands free speech communication, the method comprising:
transforming an input microphone signal into corresponding speech component signals;
applying noise reduction to the speech component signals to generate noise reduced speech component signals;
applying a non-linear function to distorted speech component signals to generate non-linear speech signals;
estimating an envelope of the speech component signals;
for voice signals of the non-linear speech signals, generating synthetic excitation signals having harmonics at predetermined frequencies, wherein the generated synthetic excitation signals are based upon the non-linear speech signals and their corresponding envelopes;
combining the estimated envelope of the speech component signals and the synthetic excitation signals to produce synthesized speech component signals; and
adaptively combining the noise reduced speech component signals and the synthesized speech component signals based on signal to noise conditions and extracted features from the input to generate enhanced speech component signals for automatic speech recognition and hands free speech communication,
wherein the enhanced speech components signals are used to recognize the input microphone signal.
2. The method according to claim 1 , wherein the speech component signals are time domain speech component signals.
3. The method according to claim 1 , wherein the speech component signals are frequency domain speech component signals.
4. The method according to claim 1 , wherein the non-linear function operates to restore distorted or missing harmonics of the distorted speech component signals.
5. The method according to claim 1 , wherein relevant features extracted from the distorted speech component signals are used for producing the synthesized speech component signals.
6. The method according to claim 1 , wherein a source-filter model is used for producing the synthesized speech component signals.
7. The method according to claim 1 , wherein the envelop of the speech component signals is estimated using a parametric model for a shape of the envelop.
8. The method according to claim 1 , wherein applying the non-linear function to distorted speech component signals includes dividing the distorted speech component signals into different frequency bands and adaptively determining an upper frequency limit for a convolution of a subset of the distorted speech component signals used to generate the non-linear speech signals based on a measured variability within each of the different frequency bands.
9. The method according to claim 1 , wherein the non-linear function is applied to the distorted speech component signals using auto-convolution of a subset of the distorted speech component signals.
10. The method according to claim 1 , wherein applying the non-linear function to distorted speech component signals includes attenuating undesired signal components from the distorted speed component signals to generate reference signals for speech reconstruction, and applying the non-linear function to the reference signals to generate the non-linear speech signals.
11. The method according to claim 1 , wherein generating the synthetic excitation signals includes performing a mathematical operation on the non-linear speech signals using their corresponding envelopes to generate the synthetic excitation signals.
12. The method according to claim 1 , wherein generating the synthetic excitation signals includes dividing the non-linear speech signals by their corresponding envelopes to generate the synthetic excitation signals.
13. The method according to claim 1 , wherein the envelop is estimated based on filter coefficients of a filter applied to the speech component signals to generate the noise reduced speech component signals.
14. The method according to claim 1 , wherein applying the non-linear function to distorted speech component signals includes applying a non-linear function to a harmonic signal to generate non-linear speech signals having sub-harmonics and super-harmonics.
15. The method according to claim 1 , wherein adaptively combining the noise reduced speech component signals and the synthesized speech component signals includes: at a first signal-to-noise ratio, replacing the noise-reduced speech component signals with the synthesized speech component signals, and at a second signal-to-noise ratio that is higher than the first signal-to-noise ratio, adaptively combining magnitudes of the noise-reduced speech component signals and the synthesized speech component signals.
16. The method according to claim 1 , wherein the extracted features from the input include variability of the speech component signals and voice/unvoiced classification of the speech component signals.
17. A speech signal processing system for automatic speech recognition and hands free speech communication employing at least one hardware implemented computer processor, the system comprising:
a signal pre-processor for transforming an input microphone signal into corresponding speech component signals;
a noise suppression module for applying noise reduction to the speech component signals to generate noise reduced speech component signals;
a non-linear operator module for applying a non-linear function to distorted speech component signals to generate non-linear speech signals;
an envelope estimation module for estimating an envelope of the speech component signals;
an excitation signal generator for generating synthetic excitation signals having harmonics at predetermined frequencies for voice signals of the non-linear speech signals, wherein the generated synthetic excitation signals are based upon the non-linear speech signals and their corresponding envelopes;
a synthesis combination module for combining the estimated envelope of the speech component signals and the synthetic excitation signals to produce synthesized speech component signals; and
a signal combination block for adaptively combining the noise reduced speech component signals and the synthesized speech component signals based on signal to noise conditions to generate enhanced speech component signals for automatic speech recognition and hands free speech communication,
wherein the enhanced speech components signals are used to recognize the input microphone signal.
18. The system according to claim 1 , wherein the non-linear function operates to restore distorted or missing harmonics of the distorted speech component signals.
19. The system according to claim 17 , wherein the speech reconstruction module uses relevant features extracted from the distorted speech component signals for producing the synthesized speech component signals.
20. The system according to claim 17 , wherein the non-linear operator module divides the distorted speech component signals into different frequency bands and adaptively determines an upper frequency limit for a convolution of a subset of the distorted speech component signals used to generate the non-linear speech signals based on a measured variability within each of the different frequency bands.