Wideband speech parameterization for high quality synthesis, transformation and quantization

• US 9,224,402 B2
• Filed: 09/30/2013
• Issued: 12/29/2015
• Est. Priority Date: 09/30/2013
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Shechtman, Slava
• Primary Examiner(s): Singh, Satwant
• Application Number:
  US14/040,765
• Publication Number:
  US 20150095035A1
• Time in Patent Office: 820 Days
• Field of Search: 704/266, 704/209, 704/233, 704/270, 704/205, 704/220, 704/264, 704/500
• US Class Current: 1/1
• CPC Class Codes: G10L 19/02 using spectral analysis, e.g. transform vocoders or subband vocoders
• CPC Class Codes: G10L 19/038 Vector quantisation, e.g. TwinVQ audio
• CPC Class Codes: G10L 19/093 using sinusoidal excitation models

Abstract

A method for speech parameterization and coding of a continuous speech signal. The method comprises dividing said speech signal into a plurality of speech frames, and for each one of the plurality of speech frames, modeling said speech frame by a first harmonic modeling to produce a plurality of harmonic model parameters, reconstructing an estimated frame signal from the plurality of harmonic model parameters, subtracting the estimated frame signal from the speech frame to produce a harmonic model residual, performing at least one second harmonic modeling analysis on the first harmonic model residual to determine at least one set of second harmonic model components, removing the at least one set of second harmonic model components from the first harmonic model residual to produce a harmonically-filtered residual signal, and processing the harmonically-filtered residual signal with analysis by synthesis techniques to produce vectors of codebook indices and corresponding gains.

Claims

1. A method for speech parameterization and coding of a continuous speech signal, comprising:

receiving a continuous speech signal representing speech recorded by at least one microphone,

dividing said continuous speech signal into a plurality of speech frames, and for each one of said plurality of speech frames;

modeling said speech frame by a first harmonic modeling to produce a plurality of harmonic model parameter values, wherein said first harmonic modeling is estimated by computing a cost function between a plurality of sine function signals and said speech frame, wherein each of said plurality of sine function signals comprises one of a plurality of harmonic frequencies, an amplitude value and a phrase value;

reconstructing an estimated frame signal from said plurality of harmonic model parameter values;

subtracting said estimated frame signal from said speech frame to produce a harmonic model residual signal;

performing at least one second harmonic modeling analysis on said first harmonic model residual to determine at least one set of second harmonic model component values;

removing said at least one set of second harmonic model component values from said first harmonic model residual signal to produce a harmonically-filtered residual signal; and

processing said harmonically-filtered residual signal with analysis by synthesis techniques to produce vectors of codebook indices and corresponding gains, and

sending said plurality of harmonic model parameter values and said codebook vector indices and corresponding gains to a speech processor configured to compute at least one of a speech transformation, a signal compression and a conversion to an audible sound output.

2. The method of claim 1, wherein said harmonic modeling is performed by using speech frame'"'"'s energy envelope estimated signal.

3. The method of claim 1, wherein said at least one set of second harmonic model component values is removed in a plurality of iterations so that during each one of said plurality of iterations the following is performed until a remaining harmonic component cost function is below a threshold:

analyzing new harmonic model of previous harmonic model residual to produce new set of harmonic model component values,

removing said new set of harmonic component values from said previous harmonic model residual to produce a new harmonic model residual for further iterations.

4. The method of claim 1, wherein said removed at least one set of harmonic component values is stored for later use during decoding of signal and reconstruction of audible output.

5. The method of claim 3, wherein said new harmonic modeling uses at least one estimated energy envelope signal.

6. The method of claim 1, wherein said speech frame is spectrally whitened prior to said first harmonic modeling, and said spectrally whitening is reversed prior to said speech coding analysis.

7. The method of claim 1, wherein said speech frame is spectrally whitened after said first harmonic modeling, and said spectrally whitening is reversed prior to said speech coding analysis.

8. The method of claim 1, wherein said harmonically-filtered residual signal is further processed to remove periodic energy envelope modulation by modeling using a sum of multiple instances of a periodic function at arbitrary frequencies taking into account the time-domain energy envelope signal estimate with imposed periodicity before analysis by synthesis coding.

9. The method of claim 8, wherein said harmonically-filtered residual signal is frequency range filtered before performing said modeling to remove only the frequency range specific periodic energy envelope modulation.

10. The method of claim 1, where said first harmonic model parameter values undergo further processing for speech transformation.

11. A method for speech parameterization and coding of a continuous speech signal, comprising:

receiving a continuous speech signal representing speech recorded by at least one microphone,

dividing said speech signal into a plurality of speech frames;

for each one of said plurality of speech frames;

modeling said speech frame by a first harmonic modeling to produce a plurality of harmonic model parameter values, wherein said first harmonic modeling is estimated by computing a cost function between a plurality of sine function signals and said speech frame, wherein each of said plurality of sine function signals comprises one of a plurality of harmonic frequencies, an amplitude value and a phrase value;

reconstructing an estimated frame signal from said plurality of harmonic model parameter values;

subtracting said estimated frame signal from said speech frame to produce a harmonic model residual signal;

removing at least one harmonic component value from said first harmonic model residual signal to produce a harmonically-filtered residual signal;

removing periodic energy envelope modulation using a second modeling of said harmonically-filtered residual signal using a sum of multiple instances of a periodic function at arbitrary frequencies taking into account the time-domain energy envelope signal estimate with imposed periodicity; and

processing said harmonically-filtered residual signal with analysis by synthesis techniques to produce vectors of codebook indices and corresponding gains, and

sending said plurality of harmonic model parameter values and said codebook vector indices and corresponding gains to a speech processor configured to compute at least one of a speech transformation, a signal compression and a conversion to an audible sound output.

12. The method of claim 11, wherein said first harmonic modeling is performed by using speech frame'"'"'s energy envelope estimated signal.

13. The method of claim 11, wherein said speech frame is spectrally whitened prior to said first harmonic modeling, and said spectrally whitening is reversed prior to said speech coding analysis.

14. The method of claim 11, wherein said harmonic model residual is spectrally whitened after said first harmonic modeling, and said spectrally whitening is reversed prior to said speech coding analysis.

15. The method of claim 11, wherein said harmonically-filtered residual signal is frequency range filtered before performing said second modeling to remove only the frequency range specific periodic energy envelope modulation.

16. The method of claim 11, where said first harmonic model parameter values undergo further processing for speech transformation.

17. An apparatus for speech parameterization and coding of a continuous speech signal, comprising:

at least one input interface for receiving and digitizing said continuous speech signal;

at least one processing unit for performing the actions of;

receiving a continuous speech signal representing speech recorded by at least one microphone,

dividing said continuous speech signal into a plurality of speech frames, and for each one of said plurality of speech frames;

modeling said speech frame by a first harmonic model to produce a plurality of frame model parameter values and harmonic model residual, wherein said first harmonic modeling is estimated by computing a cost function between a plurality of sine function signals and said speech frame, wherein each of said plurality of sine function signals comprises one of a plurality of harmonic frequencies, an amplitude value and a phrase value;

performing at least one second harmonic modeling analysis on said first harmonic model residual to remove at least one set of second harmonic model component values from said first harmonic model residual signal to produce a harmonically-filtered residual signal; and

processing said harmonically-filtered residual signal with analysis by synthesis techniques to produce vectors of codebook indices and corresponding gains, and

sending said plurality of harmonic model parameter values and said codebook vector indices and corresponding gains to a speech processor configured to compute at least one of a speech transformation, a signal compression and a conversion to an audible sound output;

at least one output interface to send said plurality of speech parameter values and codes; and

a housing for containing said at least one input interface, said at least one processing unit, and said at least one output interface, said housing being configured and suitable for the apparatus environment.

18. The apparatus of claim 17, wherein said harmonically-filtered residual signal is further processed to remove periodic energy envelope modulation using a modeling action using a sum of multiple instances of a periodic function at arbitrary frequencies taking into account the time-domain energy envelope signal estimate with imposed periodicity before analysis by synthesis coding.

19. The apparatus of claim 17, wherein said at least one input interface is any member of the group comprising:

said at least one microphone;

an analog communication interface; and

a digital communication interface.

20. The apparatus of claim 17, wherein said at least one output interface is any member of the group comprising:

a digital communication interface; and

an audio output interface.