Reduction of background noise for speech enhancement

• US 5,550,924 A
• Filed: 03/13/1995
• Issued: 08/27/1996
• Est. Priority Date: 07/07/1993
• Status: Expired due to Term

• Current Assignee: Polycom Incorporated (Plantronics, Inc.)
• Original Assignee: Picturetel Corporation (Plantronics, Inc.)
• Inventors: Helf, Brant M., Chu, Peter L.
• Primary Examiner(s): Isen, Forester W.
• Application Number:
  US08/402,550
• Time in Patent Office: 533 Days
• Field of Search: 381/94, 381/46, 381/47, 395/2.34, 395/2.35
• US Class Current: 381/94.3
• CPC Class Codes: G10L 21/0208 Noise filtering
• CPC Class Codes: G10L 21/0232 Processing in the frequency domain
• CPC Class Codes: H04B 1/1027 assessing signal quality or detecting noise/interference for the received signal
• CPC Class Codes: H04B 2001/1063 using a notch filter
• CPC Class Codes: H04S 1/002 Non-adaptive circuits, e.g. manually adjustable or static, for enhancing the sound image or the spatial distribution control circuits for electronic adaptation of the sound field H04S7/30

Abstract

Properties of human audio perception are used to perform spectral and time masking to reduce perceived loudness of noise added to the speech signal. A signal is divided temporally into blocks which are then passed through notch filters to remove narrow frequency band components of the noise. Each block is then appended to part of the previous block in a manner which avoids block boundary discontinuities. An FFT is then performed on the resulting larger block, after which the spectral components of the signal are fed to a background noise estimator. Each frequency component of the signal is analyzed with respect to the background noise to determine, within various confidence levels, whether it is pure noise or a noise-and-signal combination. The frequency band'"'"'s gain function is determined, based on the confidence levels. A spectral valley finder detects and fills in spectral valleys in the frequency component gain function, after which the function is used to modify the magnitude components of the FFT. An inverse FFT then maps the signal back from the frequency domain to the time domain to give a frame of noise-reduced signal. This signal is then multiplied by a temporal window and joined to the previous frame'"'"'s signal to derive the output.

Claims

1. An apparatus for the perceived real-time suppression of background noise in an input audio signal having components of noise comprising:

a framer for dividing the input audio signal into a sequence of audio signal frames;

a windower for producing a current windowed audio signal frame, combining all of one audio signal frame with some of the audio signal frame immediately preceding in time the one audio signal frame;

a transformer for obtaining a group of frequency spectrum components from the current windowed audio signal frame;

a noise estimator using the frequency spectrum components to produce a noise estimate of an amount of noise in the frequency spectrum components;

a noise suppression spectral modifier for producing current gain multiplicative factors based on the noise estimate and the frequency spectrum components;

a delayer for delaying, by a fixed number of frames, the frequency spectrum components of said sequence of audio signal frames to produce delayed frequency spectrum components;

a controlled attenuator for attenuating the delayed frequency spectrum components of a previous frame of said sequence based on the current gain multiplicative factors produced using said current frame to produce noise-reduced frequency components; and

an inverse transformer for converting the noise-reduced frequency components to the time domain.

2. The apparatus of claim 1 further comprising:

a post-windower for producing smoothed time-domain components, for minimizing discontinuities in the noise-reduced time-domain components;

said post-windower comprising

an overlap-and-adder for outputting a first portion of the smoothed time-domain components from said inverse transformer in combination with a previously stored portion of smoothed time-domain components from a previous frame, and for storing a remaining portion of the smoothed time-domain components from said transformer, wherein the remaining portion comprises the smoothed time-domain components not included in the first portion.

3. The apparatus of claim 1 wherein said noise suppression spectral modifier further comprises means for using previously produced gain multiplicative factors for determining current gain multiplicative factors.

4. The apparatus of claim 1 wherein said noise suppression spectral modifier comprises:

means for effecting the behavior of a fast attack, slow decay filter in said temporal domain.

5. The apparatus of claim 1 further comprising a notch filter bank for removing narrow band components of noise from said audio signal frames to produce filtered audio signal frames.

6. An apparatus for the perceived suppression of background noise in an input audio signal having components of noise comprising:

a framer for dividing the input audio signal into a sequence of audio signal frames;

a windower for producing a current windowed audio signal frame, combining all of one audio signal frame with some of the audio signal frame immediately preceding in time the one audio signal frame;

a transformer for obtaining a group of frequency spectrum components from the windowed audio signal frame;

a noise estimator using the frequency spectrum components to produce a noise estimate of an amount of noise in the frequency spectrum components;

a noise suppression spectral modifier for producing current gain multiplicative factors based on the noise estimate and the frequency spectrum components;

a delayer for delaying the frequency spectrum components of said sequence of audio signal frames to produce delayed frequency spectrum components;

a controlled attenuator for attenuating the delayed frequency spectrum components of a previous frame of said sequence based on the current gain multiplicative factors produced using said current frame to produce noise-reduced frequency components; and

an inverse transformer for converting the noise-reduced frequency components to the time domain, and

wherein the noise suppression spectral modifier further comprises;

means for using previously produced gain multiplicative factors for determining current gain multiplicative factors;

a global decision mechanism for making, for a group of the frequency spectrum components of a frame a determination as to whether that group is noise;

a local noise decision mechanism for deriving, for each frequency component of the frequency spectrum components, a confidence level for each component, whether that frequency component is a noise component;

a detector for determining, based on the confidence levels, initial gain multiplicative factors for each frequency component; and

a spreading mechanism for spectrally and temporally effecting adjustment of the initial gain multiplicative factors.

7. The apparatus of claim 6 wherein the background noise estimator produces a corresponding noise estimate for each frequency spectrum component, and wherein the local noise decision mechanism derives confidence levels based on:

(a) ratios between each of the frequency components and a corresponding noise estimate of said frequency component, and

(b) the determinations made by the global decision mechanism.

8. The apparatus of claim 7 wherein said background noise estimator further comprises:

means for comparing said ratios with selected thresholds, said thresholds being dependent upon whether a previous group was declared to be noise by said global decision mechanism.

9. The apparatus of claim 7 wherein multiplicative factors have a selected lower level value and said spreading mechanism further comprises:

means for increasing said lower level value for a frequency component of a group based upon a signal-to-noise measurement for said frequency component and said confidence level for said frequency component of said group.

10. The apparatus of claim 6 wherein the spreading mechanism adjusts the gain multiplicative factors based on the confidence levels.

11. The apparatus of claim 6 wherein said noise suppression spectral modifier further comprises:

a spectral valley filler for detecting and filling in spectral valleys in said frequency components.

12. The apparatus of claim 6 wherein said global decision mechanism further comprises:

means for comparing each frequency spectrum component with a corresponding noise estimate for said frequency component; and

means for declaring a group as a speech frame when a selected number of said spectrum components exceed, by a selected threshold amount, said corresponding noise estimate.

13. The apparatus of claim 6 wherein said spreading mechanism as like a fast attack, slow decay filter, for rapidly increasing said multiplicative factor for a component and slowly allowing said multiplicative factor to diminish to a variable lower bound for said frequency component.

14. The apparatus of claim 6 wherein said spreading mechanism comprises:

means for adjusting said multiplicative factors, due to spectral spreading, only for frequencies greater than about 500 Hz.

15. An apparatus for the perceived suppression of background noise in an input audio signal having components of noise comprising:

a transformer for obtaining frequency spectrum components from an audio signal frame derived from the input audio signal;

a detector for determining multiplicative gain factors for each frequency component;

a spreading mechanism for adjusting the multiplicative gain factors for effecting temporal and spectral spreading; and

a controlled attenuator for attenuating the frequency components in accordance with said adjusted multiplicative gain factors to derive a noise-modified spectral signal.

16. A method for reducing the perception of background noise of an input audio signal comprising the steps of:

dividing the input audio signal into a sequence of audio signal frames;

producing a windowed audio signal frame;

combining all of one audio signal frame with some of the audio signal frame immediately preceding in time the one audio signal frame to obtain a current frame;

obtaining a group of frequency spectrum components from the windowed audio signal frame;

using the frequency spectrum components to produce a noise estimate of an amount of noise in the frequency spectrum components;

producing current gain multiplicative factors based on the noise estimate and the frequency spectrum components;

delaying, by a fixed number of frames, the frequency spectrum components of said sequence to produce delayed frequency spectrum components;

attenuating the delayed frequency spectrum components of a previous frame based on the current gain multiplicative factors produced using said current frame to produce noise-reduced frequency components; and

converting the noise-reduced frequency components to the time domain.

17. The method of claim 16 further comprising the step of:

post-windowing to produce smoothed time-domain components; and

said post-windowing comprises the steps of;

outputting a first portion of the smoothed time-domain components from said step of converting in combination with a previously stored portion of smoothed time-domain components from a previous frame; and

storing a remaining portion of the smoothed time-domain components, wherein the remaining portion comprises the smoothed time-domain components not included in the first portion.

18. The method of claim 16 wherein said step of producing current gain multiplicative factors comprises the step of using previously produced gain multiplicative factors for determining said current gain multiplicative factors.

19. A method for reducing the perception of background noise of an input audio signal comprising the steps of:

dividing the input audio signal into a sequence of audio signal frames;

producing a windowed audio signal frame;

combining all of one audio signal frame with some of the audio signal frame immediately preceding in time the one audio signal frame to obtain a current frame;

obtaining a group frequency spectrum components from the current windowed audio signal frame;

using the frequency spectrum components to produce a noise estimate of an amount of noise in the frequency spectrum components;

producing current gain multiplicative factors based on the noise estimate and the frequency spectrum components;

delaying the frequency spectrum components of said sequence to produce delayed frequency spectrum components;

attenuating the delayed frequency spectrum components of a previous frame based on the current gain multiplicative factors produced using said current frame to produce noise-reduced frequency components;

converting the noise-reduced frequency components to the time domain; and

wherein the step of producing current gain multiplicative factors further comprises the steps of;

using previously produced gain multiplicative factors for determining said current gain multiplicative factors;

making, for a group of the frequency spectrum components of a frame, a determination as to whether that group is noise;

deriving, for each frequency component of the frequency spectrum components, a confidence level for each component, whether that frequency component is a noise component;

determining, based on the confidence levels, initial gain multiplicative factors for each frequency component; and

spectrally and temporally effecting adjustment of the initial gain multiplicative factors.

20. The method of claim 19 wherein said step of producing current gain multiplicative factors further comprises the step of detecting and filling-in spectral valleys in said frequency components.

21. A method for reducing the perception of background noise in an input audio signal having components of noise comprising the steps of:

obtaining frequency spectrum components from an audio signal frame derived from the input audio signal;

determining multiplicative gain factors for each frequency component;

adjusting the multiplicative gain factors for effecting temporal and spectral spreading; and

attenuating the frequency components in accordance with said adjusted multiplicative gain factors to derive a noise-modified spectral signal.