System for adaptive enhancement of speech signals

• US 8,566,086 B2
• Filed: 06/28/2005
• Issued: 10/22/2013
• Est. Priority Date: 06/28/2005
• Status: Active Grant

• Current Assignee: Blackberry Limited
• Original Assignee: QNX Software Systems Limited (Canada) (Blackberry Limited)
• Inventors: Giesbrecht, David, Hetherington, Phillip
• Primary Examiner(s): Dorvil, Richemond
• Assistant Examiner(s): ADESANYA, OLUJIMI A
• Application Number:
  US11/167,955
• Publication Number:
  US 20060293882A1
• Time in Patent Office: 3,038 Days
• Field of Search: 704/225, 704/226, 704/233, 704/240, 704/244
• US Class Current: 704/225
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence
• CPC Class Codes: G10L 21/02 Speech enhancement, e.g. noise reduction or echo cancellation reducing echo effects in line transmission systems H04B3/20; echo suppression in hands-free telephones H04M9/08
• CPC Class Codes: G10L 21/0208 Noise filtering

Abstract

A method and system for enhancing the frequency response of speech signals are provided. An average speech spectral shape estimate is calculated over time based on the input speech signal. The average speech spectral shape estimate may be calculated in the frequency domain using a first order IIR filtering or “leaky integrators.” Thus, the average speech spectral shape estimate adapts over time to changes in the acoustic characteristics of the voice path or any changes in the electrical audio path that may affect the frequency response of the system. A spectral correction factor may be determined by comparing the average speech spectral shape estimate to a desired target spectral shape. The spectral correction factor may be added (in units of dB) to the spectrum of the input speech signal in order to enhance or adjust the spectrum of the input speech signal toward the desired spectral shape, and an enhanced speech signal re-synthesized from the corrected spectrum.

Claims

1. A computer-implemented method of enhancing a frequency response of a received speech signal, the method comprising:

performing, through the use of a processor, a frequency sub-band analysis on successive overlapping windowed buffers of the received speech signal to generate a compressed dB spectrum of the received speech signal for each successive overlapping windowed buffer;

adapting a running average of a spectral shape of speech based on a current compressed dB spectrum corresponding to one of the successive overlapping windowed buffers;

subtracting, through the use of the processor, the adapted running average of the spectral shape of speech from a target spectral shape, the difference between the target spectral shape and the adapted running average of the spectral shape of speech comprising a spectral shape correction factor; and

adding, through the use of the processor, the spectral shape correction factor to the current compressed dB spectrum.

2. The method of claim 1 where the successive overlapping windowed buffers comprise Hanning windows.

3. The method of claim 1 further comprising adapting a background noise estimate for each successive overlapping windowed buffer.

4. The method of claim 3 further comprising:

determining whether signal power for each frequency sub-band of the compressed dB spectrum of each successive overlapping windowed buffer exceeds the background noise estimate by a threshold amount;

determining whether each sub-band of the compressed dB spectrum of each successive overlapping windowed buffer likely contains speech; and

adapting the running average of the spectral shape of speech for each frequency sub-band in which the signal power exceeds the background noise by at least the threshold amount and which likely contain speech.

5. The method of claim 1 where the running average of the spectral shape of speech is calculated using a first order IIR filter.

6. The method of claim 1 of further comprising re-synthesizing a speech signal from the corrected spectra corresponding to each successive overlapping windowed buffer.

7. The method of claim 1 where the target spectral shape corresponds to an ideal spectral shape of a speech signal input to a telephone system.

9. The method of claim 4 where the threshold amount varies from one frequency sub-band to the next depending on the expected noise characteristics of the system.

8. The method of cleaning 1 where the target spectral shape corresponds to an ideal spectral shape of a speech signal input to a voice recognition system.

10. A system for enhancing the frequency response of a speech signal comprising:

a microphone for capturing a speech signal;

an A/D converter for converting the speech signal into a digital speech signal; and

a processor adapted to continuously update a running average of a spectral shape of the speech signal received at the microphone, to subtract the continuously updated running average of the spectral shape of the speech signal from a target spectral shape, the difference between the target spectral shape and the adapted running average of the spectral shape of speech comprising a speech spectral shape correction factor, and to adjust the speech signal using the speech spectral shape correction factor.

11. The system of claim 10 further comprising an application configured to utilize the speech signal having a spectrum adjusted by the processor based on differences between the continuously updated average spectral shape of the speech signal and the target spectral shape.

12. The system of claim 11 where the application is a hands free telephone system.

13. The system of claim 11 where the application is a speech recognition system.

14. A computer-implemented method of enhancing a frequency response of a speech signal comprising:

performing, through the use of a processor, a frequency sub-band analysis on successive overlapping windowed buffers of the speech signal to generate a compressed dB spectrum of the received speech signal for each successive overlapped windowed buffer;

generating, through the use of the processor, a background noise estimate across the frequency sub-bands;

generating, through the use of the processor, a background noise spectral shape correction factor by subtracting the background noise estimate from a target background noise spectral shape; and

adding, through the use of the processor, the background noise spectral shape correction factor to a spectrum corresponding to one of the successive overlapping windowed buffers.

15. The method of claim 14 where the successive overlapping windowed buffers comprise Hanning windows.

16. The method of claim 14 further comprising re-synthesizing a speech signal from the corrected spectra corresponding to each successive overlapping windowed buffer.

17. The method of claim 14 where the target background noise spectral shape corresponds to smooth broad band background noise.

18. A computer-implemented method of enhancing a frequency response of a speech signal comprising:

performing, through the use of a processor, a frequency sub-band analysis on successive overlapping windowed buffers of said speech signal to generate a compressed dB spectrum of the received speech signal for each successive overlapped windowed buffer;

adapting a running average of a spectral shape of speech based on a current compressed dB spectrum corresponding to one of the successive overlapping windowed buffers;

subtracting, through the use of the processor, the adapted running average of the spectral shape of speech from a target spectral shape, the difference between the target spectral shape and the adapted running average of the spectral shape of speech comprising a spectral shape correction factor;

generating, through the use of the processor, a background noise estimate across the frequency sub-bands;

calculating, through the use of the processor, a background noise spectral shape correction factor corresponding to a difference between the background noise estimate and a target background noise spectral shape;

calculating, through the use of the processor, an overall spectral shape correction factor based on the speech spectral shape correction factor and the background noise spectral shape correction factor; and

adding, through the use of the processor, the overall spectral shape correction factor to a spectrum corresponding to one of the successive overlapping windowed buffers,

where the step of calculating, through the use of the processor, an overall spectral correction factor comprises inversely weighting the speech spectral shape correction factor and the background noise spectral shape correction factor according to a long term SNR estimate.