The present invention relates generally to noise suppression and in particular, to background noise suppression for speech within a speech coding system.
Cellular telephones, speaker phones, and various other communication devices utilize background noise suppression to enhance the quality of a received signal. In particular, the presence of acoustic background noise can substantially degrade the performance of a speech communication system. The problem is exacerbated when a narrow-band speech coder is used in the communication link since such coders are tuned to specific characteristics of clean speech signals and handle noisy speech and background noise rather poorly.
A simplified block diagram of the basic noise suppression system 100 is shown in
Although the above technique does serve to reduce the background noise, it was observed that background noise could produce annoying artifacts when entering the transition region of the gain curve since background noise will have short-term SNR fluctuations around the 0 dB origin since the channel noise energy estimator smoothes the energy via low-pass filtering. As a result, the channel energy estimate moves quicker than the respective noise energy estimate, and the short-term fluctuations in SNR (and subsequently, gain) cause “waterfall” or “swirling” artifacts. To circumvent this problem, prior-art techniques have proposed a method by which the channel SNR estimate is modified to include a process that 1) detects spurious activity in the transition region, and 2) sets the channel SNR back to zero when the signal is spurious. This method is illustrated in
A problem exists in that in order to detect that a channel SNR is “spurious”, it is required that only “some” of the channel SNRs enter into the transition region. This is fine for stationary noises that have uncorrelated frequency components (e.g., wind noise in a car), but in cases where the frequency components are correlated (e.g., office noise, interfering talkers, impulsive noise, etc.), the method cannot discriminate between non-stationary background noise and speech.
More recent efforts to improve Noise Suppression performance have focused on a “variable attenuation” concept. In order to alleviate these unpleasant effects, the algorithm was modified to adaptively reduce the amount of noise reduction during severe SNR conditions.
While this method has proven to be effective in low SNR environments, it does not address the ongoing problem of non-stationary, impulsive type noises. Thus a need exists to improve performance of prior-art noise suppression systems for non-stationary noises, while maximizing the benefits associated with the variable attenuation concept.
To address the above-mentioned need, a method and apparatus for noise suppression is described herein. In accordance with the preferred embodiment of the present invention a channel gain is additionally controlled based on a degree of variability of the background noise. The noise variability estimate is used in conjunction with the variable attenuation concept to produce a family of gain curves that are adaptively suited for a variety of combinations of long-term peak SNR and noise variability. More specifically, a measure of the variability of the background noise is used to provide an optimized threshold that reduces the occurrence of non-stationary background noise entering into the transition region of the gain curve.
Utilizing this technique for adjusting the gain of a speech-plus-noise signal results in improved performance over prior-art noise suppression systems for non-stationary noises, while maximizing the benefits associated with the variable attenuation concept
The present invention encompasses an apparatus comprising a noise variability estimator determining an amount of variability of background noise in a speech-plus-noise signal, and a channel gain generator adjusting a gain applied to the speech-plus-noise signal based on the amount of variability in the background noise.
The present invention additionally encompasses an apparatus for noise suppression. The apparatus comprises a channel signal energy estimator for estimating a total energy of a speech-plus-noise signal, a noise energy estimator for estimating a noise energy of the speech-plus-noise signal, a channel signal-to-noise (SNR) estimator having the noise energy estimate and the total energy estimate as an input and outputting an SNR estimate of the speech-plus-noise signal, a SNR variability estimator for estimating the SNR variability of the speech-plus-noise signal, and a channel gain generator for attenuating the speech-plus-noise signal based on the SNR variability of the speech-plus-noise signal.
The present invention additionally encompasses a method for noise suppression, the method comprising the steps of estimating an amount of variability in background noise in a speech-plus-noise signal and adjusting a gain applied to the speech-plus noise signal based on the amount of variability in the background noise.
Turning now to the drawings, wherein like numerals designate like components,
In order to facilitate the use of the present invention, a method for measuring the variability of the background noise is needed. One such method utilizes the variability of SNR, and has been developed for use in a Voice Activity Detection (VAD) algorithm, as disclosed in U.S. patent application Ser. No. 09/293,448, entitled A
Once the variation in channel noise is determined, in a first embodiment of the present invention the channel gain generator adjusts the attenuation of the speech/noise FFT signal as illustrated in
If, on the other hand, it is expected that low SNR, high variability noise may be encountered in practice, the gain curve of
In the preferred embodiment of the present invention a combination of the first two embodiments is implemented, as shown in
γDB(i)=μg(σq″(i)−σth)+γn; 0≦i<Nc, (1)
where μg is the gain slope, σq″ (i) is the channel SNR for channel i, σth is the SNR threshold, γn is the minimum overall gain (e.g., −13 dB), and Nc is the number of frequency channels. The result of this equation is further constrained to be within γn≦γDB(i)≦0.
From these equations it can be seen that any channel SNR σq″ (i) below the SNR threshold σth will result in the minimum channel gain γn being applied to that channel i. As the channel SNR exceeds the SNR threshold, the transition region of the gain curve is entered, until a point at which the gain is limited to 0 dB. In the preferred embodiment of the present invention, the SNR threshold and minimum gain are allowed to vary as a function of variability of the background noise σn, (i.e., γn→γ(σn), σth→σth(σn)). From this, the equation given in (1) can be modified in accordance with the current invention as:
γDB(i)=μg(σq″(i)−σth(σn))+γ(σn);0≦i<Nc, (2)
which is subsequently limited by: γ(σn)≦γDB(i)≦0. These equations reflect the gain characteristic as shown in
Continuing, once the variability in the SNR is determined, it is output to channel gain generator 502 along with the SNR estimate where the channel gain is adjusted accordingly (step 1013) based on the amount of variability in the background noise of the speech/noise signal. Finally, at step 1015 an inverse FFT of the attenuated signal takes place by IFFT circuitry 102.
As described above, in a first embodiment of the present invention, the channel gain is adjusted based on the SNR and the SNR variation such that voice/noise signal is attenuated a first amount when the SNR variation is high, otherwise the signal is attenuated a second amount as illustrated in
Additionally, in the second embodiment a maximum attenuation of the signal is additionally based on the variability in background noise (SNR variation) as shown in
Finally, in a third, and preferred embodiment a combination of the first two embodiments takes place. In particular, for the same SNR value the attenuation of the signal will vary, based on the variability of the SNR, with higher amounts of attenuation taking place in situations with higher SNR variation levels. Additionally, the maximum amount of attenuation will vary based on the SNR variability, with a more attenuation of signals with higher SNR variability taking place.
While the invention has been particularly shown and described with reference to a particular embodiment, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention. It is intended that such changes come within the scope of the following claims.
Number | Name | Date | Kind |
---|---|---|---|
4627103 | Fukuhara | Dec 1986 | A |
4628529 | Borth et al. | Dec 1986 | A |
4630304 | Borth et al. | Dec 1986 | A |
4811404 | Vilmur et al. | Mar 1989 | A |
5323337 | Wilson et al. | Jun 1994 | A |
5544250 | Urbanski | Aug 1996 | A |
5659622 | Ashley | Aug 1997 | A |
5937377 | Hardiman et al. | Aug 1999 | A |
6104993 | Ashley | Aug 2000 | A |
6415253 | Johnson | Jul 2002 | B1 |
6453289 | Ertem et al. | Sep 2002 | B1 |
6453291 | Ashley | Sep 2002 | B1 |
6718301 | Woods | Apr 2004 | B1 |
6862567 | Gao | Mar 2005 | B1 |
Number | Date | Country | |
---|---|---|---|
20040052384 A1 | Mar 2004 | US |