Speech coding apparatus, linear prediction coefficient analyzing apparatus and noise reducing apparatus

• US 6,205,421 B1
• Filed: 12/30/1999
• Issued: 03/20/2001
• Est. Priority Date: 12/19/1994
• Status: Expired due to Term

• Current Assignee: Panasonic Intellectual Property Corporation of America (Panasonic Holdings Corporation)
• Original Assignee: Matsushita Electric Industrial Company Limited (Panasonic Holdings Corporation)
• Inventors: Morii, Toshiyuki
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): CHAWAN, VIJAY B
• Application Number:
  US09/475,249
• Time in Patent Office: 446 Days
• Field of Search: 704/226, 704/227, 704/228, 704/229
• US Class Current: 704/226
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/16 using artificial neural networks
• CPC Class Codes: G10L 19/06 Determination or coding of the spectral characteristics, e.g. of the short-term prediction coefficients
• CPC Class Codes: G10L 19/09 Long term prediction, i.e. removing periodical redundancies, e.g. by using adaptive codebook or pitch predictor
• CPC Class Codes: G10L 19/18 Vocoders using multiple modes
• CPC Class Codes: G10L 2021/02168 the estimation exclusively taking place during speech pauses
• CPC Class Codes: G10L 21/0208 Noise filtering
• CPC Class Codes: G10L 21/0216 characterised by the method used for estimating noise
• CPC Class Codes: G10L 25/06 the extracted parameters being correlation coefficients
• CPC Class Codes: G10L 25/12 the extracted parameters being prediction coefficients
• CPC Class Codes: G10L 25/27 characterised by the analysis technique

Abstract

A sample speech is analyzed by a speech analyzing unit to obtain sample characteristic parameters, and a coding distortion is calculated from the sample characteristic parameters in each of a plurality of coding modules. The sample characteristic parameters and the coding distortions are statistically processed by a statistical processing unit to obtain a coding module selecting rule. Thereafter, when a speech is analyzed by the speech analyzing unit to obtain characteristic parameters, an appropriate coding module is selected by a coding module selecting unit from the coding modules according to the coding module selecting rule on condition that a coding distortion for the characteristic parameters is minimized in the appropriate coding module. Thereafter, the characteristic parameters of the speech are coded in the appropriate coding module, and a coded speech is obtained. When the coded speech is decoded, a reproduced speech is obtained. Accordingly, because an appropriate coding module can be easily selected from a plurality of coding modules according to the coding module selecting rule, any allophone occurring in a reproduced speech can be prevented at a low calculation volume.

Claims

1. A noise reducing apparatus, comprising:

speech signal receiving means for receiving a plurality of frames of analog speech signals in which a noise exists and converting the frames of analog speech signals into a plurality of frames of digital speech signals one after another;

Fourier transforming means for performing a discrete Fourier transformation for each of the frames of digital speech signals obtained by the speech signal receiving means and producing an input spectrum and a phase spectrum corresponding to each of the frames of digital speech signals for each of frequency values;

noise assuming means for selecting a particular input spectrum having a minimum value from among a current input spectrum, produced by the Fourier transforming means, corresponding to a current frame of digital speech signals and a predetermined number of past input spectra, produced by the Fourier transforming means, corresponding to past frames of digital speech signals preceding to the current frame and assuming the particular input spectrum as a noise spectrum corresponding to the current frame of digital speech signals for each of the frequency values;

noise reducing degree determining means for determining a degree of a noise reduction according to each of the frames of digital speech signals obtained by the speech signal receiving means;

noise reducing means for adjusting a value of each of the noise spectra assumed by the noise assuming means according to the degree of the noise reduction determined by the noise reducing degree determining means to produce an adjusted noise spectrum having an adjusted value corresponding to the current frame of digital speech signals for each of the frequency values, subtracting the adjusted noise spectrum from the current input spectrum produced by the Fourier transforming means for each of the frequency values to reduce the noise existing in the current frame of digital speech signals, and producing a noise-reduced input spectrum corresponding to the current input spectrum for each of the frequency values; and

inverse Fourier transforming means for performing an inverse Fourier transformation for the noise-reduced input spectra produced by the noise reducing means according to the phase spectra, produced by the Fourier transforming means, corresponding to the current frame of digital speech signals, producing a current frame of first-order output signals corresponding to the current frame of digital speech signals, and outputting a plurality of frames of first-order output signals corresponding to the frames of analog speech signals received by the speech signal receiving means, one after another as a plurality of frames of output signals.

2. A noise reducing apparatus according to claim 1, further comprises:

linear prediction coefficient analyzing means for performing a linear prediction analysis for each of the frames of digital speech signals obtained by the speech signal receiving means and producing a linear prediction coefficient for each of the frames of digital speech signals; and

spectrum emphasizing means for emphasizing a spectrum of each frame of the output signals output from the inverse Fourier transforming means, the frames of output signals emphasized being output.

3. A noise reducing apparatus according to claim 2 in which the emphasis of the spectrum of each of the output signals is performed by using a very emphasizing filter, a high frequency emphasizing filter and a power emphasizing filter according to the linear prediction coefficient produced by the linear prediction coefficient analyzing means.

4. A noise reducing apparatus according to claim 1, further comprises:

spectrum compensating means for looking for a particular noise-reduced input spectrum, produced by the noise reducing means, having a value lower than a predetermined value and compensating the particular noise-reduced input spectrum for an excessive subtraction of the adjusted noise spectrum, the particular noise-reduced input spectrum compensated being inversely transformed by the inverse Fourier transforming means.

5. A noise reducing apparatus according to claim 2, further comprises:

waveform matching means for matching a waveform of a current frame of output signals output from the inverse Fourier transforming means to a waveform of a preceding frame of output signals, output from the inverse Fourier transforming means, preceding the current frame, a plurality of frames of output signals matched being output one after another.

6. A noise reducing apparatus according to claim 2 in which each of the input spectra produced by the Fourier transforming means is a pseudo-amplitude spectrum obtained by adding an absolute value of a real component of a complex spectrum obtained according to the discrete Fourier transformation and an absolute value of an imaginary component of the complex spectrum.

7. A noise reducing apparatus according to claim 2, further comprising:

noise spectrum storing means for storing one noise spectrum having a smallest value among a plurality of noise spectra, assumed by the noise assuming means, corresponding to a particular frame and a plurality of particular frequency values as a representative noise spectrum corresponding to the particular frame and the particular frequency values, a value of the representative noise spectrum stored in the noise spectrum storing means being adjusted by the noise reducing means to produce an adjusted noise spectrum having an adjusted value.

8. A noise reducing apparatus according to claim 2, further comprising:

noise spectrum storing means for storing a first noise spectrum, assumed by the noise assuming means, corresponding to the current and a first predetermined number of first past frames of digital speech signals as a first-place candidate for a particular frequency value, a first continuation number equal to the number of the first past frames, a second noise spectrum, assumed by the noise assuming means, corresponding to the current and a second predetermined number of second past frames of digital speech signals as a second-place candidate for the particular frequency value, and a second continuation number equal to the number of the second past frames, the first noise spectrum being adjusted by the noise reducing means to produce an adjusted noise spectrum having an adjusted value in cases where the first continuation number is equal to or smaller than a referential continuation number, and the first noise spectrum being abandoned to set the second noise spectrum as the first-place candidate and the second noise spectrum set as the first-place candidate being adjusted by the noise reducing means to produce an adjusted noise spectrum having an adjusted value in cases where the first continuation number is larger than the referential continuation number.

9. A noise reducing apparatus according to claim 1 in which the noise reducing degree determining means comprises:

stabilizing coefficient storing means for storing a stabilizing coefficient indicating the stability of the noise spectra assumed by the noise assuming means for each of the digital speech signals; and

noise reducing coefficient adjusting means for adjusting a noise reducing coefficient as the degree of the noise reduction according to the stabilizing coefficient stored in the stabilizing coefficient storing means and reducing the stabilizing coefficient to indicate a higher stability of the noise spectra assumed by the noise assuming means each time a frame of digital speech signals is obtained by the speech signal receiving means.

10. A noise reducing apparatus according to claim 9 in which a value of each of the noise spectra assumed by the noise assuming means is multiplied by the noise reducing coefficient adjusted by the noise reducing coefficient adjusting means in the noise reducing means to produce an adjusted noise spectrum having an adjusted value.

11. A noise reducing apparatus according to claim 9 in which a noise power is calculated by the noise assuming means by adding the noise spectra assumed by the noise assuming means together, the stabilizing coefficient stored in the stabilizing coefficient storing means is compulsorily set to a high value in cases where the noise power is lower than a referential noise power, and the adjusted value of each of the adjusted noise spectra produced according to the degree of the noise reduction which is adjusted according to the stabilizing coefficient having the high value is equal to the value of each of the noise spectra.

12. A noise reducing apparatus according to claim 9, further comprising:

previous spectrum storing means for storing a plurality of previous input spectra for all frequency values and storing a plurality of current input spectra at a current frame produced by the Fourier transforming means in place of the previous input spectra as other previous input spectra in cases where the stabilizing coefficient stored in the stabilizing coefficient storing means indicates that a current noise spectrum at the current frame assumed by the noise assuming means is unstable; and

spectrum compensating means for selecting a particular input spectrum having a lower value between a current input spectrum for a particular frequency value produced by the Fourier transforming means and a particular previous input spectrum for the particular frequency value stored in the previous spectrum storing means in cases where a particular current noise-reduced input spectrum for the particular frequency value has a negative value and outputting the particular input spectrum as the particular current noise-reduced input spectrum compensated to the inverse Fourier transforming means, the particular input spectrum being replaced with the particular previous input spectrum stored in the previous spectrum storing means.

13. A noise reducing apparatus according to claim 9, further comprising:

previous spectrum storing means for storing a plurality of previous input spectra for all frequency values, storing a plurality of current input spectra at a current frame produced by the Fourier transforming means in place of the previous input spectra as other previous input spectra in cases where the stabilizing coefficient stored in the stabilizing coefficient storing means is higher than a particular value to indicate that a current noise spectrum at the current frame assumed by the noise assuming means is unstable, and storing a previous sound power obtained by adding values of the previous input spectra; and

spectrum compensating means for selecting a particular input spectrum having a lower value between a current input spectrum for a particular frequency value produced by the Fourier transforming means and a particular previous input spectrum for the particular frequency value stored in the previous spectrum storing means in cases where a particular current noise-reduced input spectrum for the particular frequency value has a negative value, regarding the particular input spectrum as an output spectrum for the particular frequency value, regarding the particular current noise-reduced input spectrum as an output spectrum for the particular frequency value in cases where the particular current noise-reduced input spectrum has a zero value or a positive value, adding a plurality of output spectra for all frequency values together to obtain an output sound power, replacing the previous input spectrum stored in the previous spectrum storing means with the output spectra, replacing the previous sound power stored in the previous spectrum storing means with the output power as another previous sound power in cases where the stabilizing coefficient stored in the stabilizing coefficient storing means is higher than the particular value, calculating a new previous sound power from the previous sound power stored in the previous spectrum storing means and the output power in cases where the output power is lower than a multiplied previous sound power obtained by multiplying the previous sound power by a first factor and the previous sound power is lower than a multiplied noise power obtained by multiplying a noise power equal to a sum of values of the noise spectra for all frequency values assumed by the noise assuming means by a second factor, calculating a power ratio by dividing the new previous sound power by the output power, producing a plurality of new output spectra by multiplying the output spectra by the power ratio, replacing the previous sound power stored in the previous spectrum storing unit with the new previous sound power, and outputting the new output spectra to the inverse Fourier transforming unit as the noise-reduced input spectra for all frequency values.

14. A noise reducing apparatus according to claim 1, further comprising:

previous spectrum storing means for storing a plurality of previous input spectra for all frequency values;

spectrum compensating means for selecting a particular input spectrum having a lower value between a current input spectrum for a particular frequency value produced by the Fourier transforming means and a particular previous input spectrum for the particular frequency value stored in the previous spectrum storing means in cases where a particular current noise-reduced input spectrum for the particular frequency value has a negative value and outputting the particular input spectrum as the particular current noise-reduced input spectrum compensated to the inverse Fourier transforming means, the particular input spectrum being replaced with the particular previous input spectrum stored in the previous spectrum storing means.

15. A noise reducing apparatus according to claim 1, further comprises:

pre-read waveform storing means for storing a head portion of a current frame of output signals output from the inverse Fourier transforming means; and

waveform matching means for receiving the current frame of output signals and a head portion of a succeeding frame of output signals output from the inverse Fourier transforming means, deforming a waveform of the current frame received according to the head portion of the current frame stored in the pre-read waveform storing means to match the waveform of the current frame to a waveform of a preceding frame of output signals, transmitting the head portion of the succeeding frame of output signals to the pre-read waveform storing means to replace the head portion of the current frame stored in the pre-read waveform storing means with the head portion of the succeeding frame of output signals, and outputting the matched current frame of output signals.