WO2009096958A1

WO2009096958A1 - Noise suppressor system and method

Info

Publication number: WO2009096958A1
Application number: PCT/US2008/052459
Authority: WO
Inventors: Tomas Fritz Gaensler; James A. Johanson; Peter Kroon; Min Liang
Original assignee: Agere Systems Inc.
Priority date: 2008-01-30
Filing date: 2008-01-30
Publication date: 2009-08-06

Abstract

A method and apparatus for a noise suppression system includes, a first noise suppressor algorithm that processes an audio input signal to provide first noise suppression gains, a second noise suppressor algorithm that processes the input signal to provide second noise suppression gains, a gain combiner tuned to the outputs of the noise suppression gains and combining the first and second noise suppression with frequency sub-bands of the input signal gains in a single processing channel, and a synthesis filter band module in the single processing channel to form a time-domain output signal of reduced noise relative to the input signal.

Description

NOISE SUPPRESSOR SYSTEM AND METHOD.

FIELD OF THE INVENTION

[0001] The present invention relates to a method and system for applying different noise suppression algorithms for noise suppression of a voice signal.

BACKGROUND

[0002] A system for noise suppression of a voice signal is based on time domain filtering of the voice signal or is based on spectral subtraction, wherein a spectral estimate or a spectral prediction of the noise wave forms are subtracted from the voice signal. [0003] US 6,035,048 discloses a stationary noise suppressor system based on spectral subtraction. The audio (voice) signal is analyzed by an analysis algorithm and divided into frequency sub-bands of noise sub-bands and voice sub-bands. A signal gain function is applied to the sub-bands to obtain suppression gains for the noise sub-bands. The respective voice sub-bands are synthesized into an enhanced signal bandwidth in which the noise-sub- band gains are attenuated. [0004] US 2003/0147538 Al discloses a spatial noise suppressor system based on spectral subtraction for attenuating directional noise. Various microphones in an array of two or more microphones receive separate inputs of a voice signal. The directional noise will vary when received by the various microphones, due to propagation speed characteristics and/or phase change characteristics. Such characteristics differ from corresponding characteristics of stationary noise inputs. A spatial noise suppression algorithm is applied to the voice signal arrays to segregate directional noise sub-bands from the desired voice sub- bands. Then, a synthesizer sums the powers of the directional noise sub-bands, and the powers of the input voice signal arrays. Any high-number signals are attenuated to suppress directional noise on an output of the voice signal. [0005] US 2007/0150268 Al discloses a system for noise suppression of a voice signal based on spectral subtraction. An array of three microphones provide three voice signal inputs to a stationary noise suppression module and a spatial noise reduction module. The voice signal is converted to sub-bands by a beam- former module, and the stationary noise suppressor removes any residual ambient or instrumental stationary noise. Thus, a first processing channel is required for removal of the stationary noise. Another processing channel is required for removal of directional noise. At the noise suppression module, the three voice signal inputs at the microphones are decomposed into two quantities of spatial direction information. A first quantity represents three microphone inputs converted to three sets of sub-bands. A second quantity is based on summed combinations of the three sets of sub-bands. Spatial noise attenuation gains are derived by a process of beam forming. The synthesizer processes the sub-bands of the voice signal for attenuation of the fixed spatial noise gains. An output of the synthesizer is free of the of the spatial noise and is combined with the stationary noise free output from the stationary noise suppressor module. [0006] Prior to the invention, a noise suppressor system that combines different noise suppression systems has required two processing channels; a first processing channel to remove stationary noise, and a second processing channel to remove directional noise. Even when combining a stationary noise suppressor and a directional noise suppressor as in US 2003/0147538 Al, a first processing channel is required in which a stationary noise suppressor removes stationary noise and an output is produced by the stationary noise suppressor, and a second processing channel is required in which a directional noise suppressor removes directional noise for synthesis with the output of the stationary noise suppressor. A disadvantage of combining different noise suppressor systems is the need for processor capacity in chip architecture to remove noise in one processing channel, and the need for further processor capacity in chip architecture to remove noise in another processing channel. It would be desirable to combine different noise suppressor systems, while reducing the processor capacity requirements in chip architecture to attenuate noise with the different noise suppressor systems.

SUMMARY OF THE INVENTION

[0007] The invention refers to a method and apparatus combining different noise suppressors of an input signal to provide first and second sub-bands of the input signal and corresponding first and second noise attenuation gains, and combining the first and second attenuation gains with the first sub-bands of the input signal in a gain combiner in a single processing channel, and synthesizing the first and second noise attenuation gains with the first sub-bands to provide an output signal having less noise than the input signal. The invention advantageously eliminates excessive signal processing capacity used in different processing channels to remove noise in one processing channel and to remove noise in another processing channel. The invention advantageously eliminates a need for signal processing capacity to synthesize noise suppression gains with the second sub-bands of the input signal. [0008] An embodiment of the invention refers to a method and apparatus combining different noise suppressors of an input signal to provide a first processing channel in which first sub-bands of the input signal are processed to provide a first output comprising first noise attenuation gains, and further to provide a second processing channel in which second sub-bands of the input signal are processed to provide a second output comprising second noise attenuation gains, and to combine the second processing channel and the first processing channel prior to synthesis of the first and second noise attenuation gains with one of, the first sub-bands or the second sub-bands, to form an output signal having reduced noise relative to the input signal. [0009] According to an embodiment of the invention, a method includes; (a) analyzing an input signal into first sub-bands in a first processing channel and analyzing the input signal into second sub-bands in a second processing channel, (b) applying a first noise suppressor algorithm to the first sub-bands to calculate first noise attenuation gains (c) applying a second noise suppressor algorithm to the second sub-bands to calculate second noise attenuation gains, (d) combining the first noise suppression gains and the second noise suppression gains and the first sub-bands in a corresponding one of the first and second processing channels; (e) synthesizing the first and second noise suppression gains to attenuate noise in the input signal; and (f) outputting a noise-attenuated signal. [0010] According to another embodiment of the invention, a gain combiner is frequency tuned to outputs comprising first and second noise suppression gains of the first and second processing channels; and a processing module synthesizes the first and second noise suppression gains with sub-bands of the input signal in one of, the first processing channel or the second processing channel.

BRIEF DESCRIPTION OF THE DRAWINGS [0011] Embodiments of the invention will now be described by way of example with reference to the accompanying drawings.

[0012] Fig. 1 is a block diagram of a noise suppressor system having multiple exemplary noise suppressors.

[0013] Fig. 2 is a block diagram of a core NSl Module. [0014] Fig. 3 is a graph of suppression gain compared to raw gain.

[0015] Fig. 4 is a block diagram of core modules of another noise suppressor system with another embodiment of two exemplary noise suppressors.

[0016] Fig. 5 is a block diagram of a signal transmission system. DETAILED DESCRIPTION

[0017] Fig. 1 discloses an embodiment of a noise suppressor system 100, which processes an input, noisy voice signal X(n) to attenuate noise frequency sub-bands that may be present in the input voice signal. The noise suppressor system 100 comprises a combination of noise suppressors NSl and NS2. Each of the noise suppressors NSl and NS2 comprises either a stationary noise suppressor or a directional noise suppressor, for example. [0018] A stationary noise suppressor attenuates omnidirectional or stationary noise, for example, ambient noise (white noise) from internal circuitry operations or resulting from a person breathing into one or more microphones. An example wherein a voice (audio) signal is analyzed into sub-band signals, includes, but is not limited to an analysis algorithm disclosed by US 6,035,048 to Diethorn, or a beamformer module disclosed by US 2007/0150268 Al of Acero et al., or a single channel Wiener filter disclosed by US 2003/0147538 Al ofElko. [0019] A directional noise suppressor attenuates directional noise such as, wind noise and background noise having such dynamics as varied azimuth direction with constant or varying decibels (dB). Further, the stationary noise suppressor is unable to calibrate the noise-to-speech ratio for improving speech intelligibility. A directional noise suppressor, for example, the noise suppressor NS2, requires an array of at least two microphones to receive the voice signal. Examples of a directional noise suppressor includes, but is not limited to a spatial noise reduction module of US 2007/0150268 Al of Acero et al. or a multi-channel Weiner filter disclosed by US 2003/0147538 Al of Elko.

[0020] There is always a trade -off between the amount of noise suppression and voice quality and voice attenuation. Increasing the level of noise suppression progressively introduces more voice degradation. Moreover, processing a single signal sequence by a single noise suppressor, for example, the noise suppressor NS 1 , is not effective to suppress noise as would a combination of noise suppressors NSl, NS2, which can apply a combination of noise suppression algorithms, wherein the combination performs better than a single noise suppression algorithm. Thus, the invention combines multiple noise suppressors to obtain more effective noise suppression than can be obtained by a single noise suppressor. The combined noise suppressors NS 1 , NS2 advantageously convert the input voice signal to frequency domain sub-bands by applying different sub-band converters, including but not limited to an analysis filter bank as disclosed by US 6,035,048 to Diethorn, or a beam former module as disclosed by US 2007/0150268 A, or a multi-channel Wiener filter as disclosed by US2003/0147538 Al to Elko. In the case of a beam former or a Wiener filter, the corresponding noise suppressor NSl, NS2 analyzes the output of the beam former or Wiener filter to frequency domain sub-bands. The input voice signal is analyzed into sub-bands to detect the presence of noise sub-bands.

[0021] The invention compensates for errors in sub-band analysis by using more than one signal-to-sub-band analyzers in respective processing channels. An embodiment of the invention comprises a combination of multiple noise suppressors NSl and NS2, which convert the input voice signal to sub-bands by applying different signal-to-sub-band analyzers in respective processing channels. [0022] Embodiments of the invention comprise, a combination of stationary and directional noise suppressors, or a combination of stationary noise suppressors, or a combination of directional noise suppressors.

[0023] Another embodiment of the invention comprises a combination of multiple noise suppressors NS 1 and NS2 applying different noise suppression algorithms to respective voice signal sub-bands in respective processing channels, and combining their outputs in a single processing channel prior to synthesis with voice signal sub-bands of the single processing channel.

[0024] According to embodiments of the invention, a combination of noise suppressors NSl and NS2 share one or more of the microphone inputs X(n), Xl(n) and X2(n) of the input signal to be analyzed for the presence of noise frequency sub-bands. Accordingly, only two microphone inputs Xl(n) and X2(n) are used in an embodiment of the invention. Another embodiment of the present invention provides a single microphone for splitting an input voice signal into one or more input signals, such as, X(n), Xl (n) and X2(n). [0025] Each of the noise suppressors NSl and NS2 samples the input (voice) signal by segmenting the signal into frames of fixed millisecond durations, with the frames shifted with a millisecond frame shift to provide buffered, moving frames or windows for sampling. The moving frames of input wave forms are converted to the frequency domain, and correlated into different frequency bins, by applying a waveform transform algorithm including, but not limited to a Fast Fourier Transform (FFT). Thereby, the input wave forms are converted to the frequency domain, and are correlated into one or more frequency sub- band components stored in frequency bins of the core memory blocks. The sub-band components include noise frequency sub-band components when noise is present on the input signal. If there is no noise on the input signal the algorithm provides a unity noise suppression gain. The noise suppressor NS 1 , NS2 generates the suppression gain for attenuation of the noise frequency sub-band components according to the following description.

[0026] Fig. 1 discloses a first exemplary processor module 100 of a first exemplary noise suppressor NSl having core modules to provide noise suppression. Fig. 1 discloses a second exemplary processor module 110 having core modules of another exemplary noise suppressor NS2 to provide additional noise suppression that supplements the noise suppression provided by the first exemplary noise suppressor NS 1. An analog baseband (ABB) chip is capable of supporting a two-channel input of NS2. The second exemplary processor module 110 processes the noisy voice signal input in a corresponding AFB module 112 and a corresponding NS2 Core module 114. The corresponding AFB module 112 performs corresponding operations similar to those described with reference to the AFB module 102 of the first exemplary noise suppressor NSl. The corresponding NS2 Core module 114 performs corresponding operations similar to those described with reference to the NSl Core module 104 of the first exemplary noise suppressor NS 1. [0027] Fig. 1 discloses the block diagram modules of the exemplary NSl algorithm and NS2 algorithm. The noise suppressor NS 1 comprises a first processing channel having three modules in cascade: an analysis filter bank (AFB) module 102, a NSl Core module 104 and a synthesis filter bank (SFB) module 106. The exemplary noise suppressor NS2 comprises a second processing channel having three modules in cascade: a LMS based Calibration module 116, an analysis filter bank (AFB) module 112, and a core NS2 module 114. In the following description, the rationale and functionality of each module will be discussed in detail. The Analysis Filter Bank (AFB) module and the Synthesis Filter Bank (SFB) Module in a corresponding processing channel are paired operations. [0028] The analysis and synthesis filter banks of the modules are based on the weighted overlap-add technique presented in R. Crochiere & L. Rabiner, "Multirate Digital Signal Processing," Prentice Hall, 1983.

[ 0029 ] The analysis filter banks of the AFB module 102, 112 analyze the time- domain input signal X(n) into signal sub-band components X(k), Ys(n) and Yd(n) in the frequency domain. According to an alternative embodiment of the invention, the AFB modules 102, 112 analyze the input signal X(n) from a beam-former as in US 2007/0150268 Al of Acero et al, or from a multi-channel Weiner filter disclosed by US 2003/0147538 Al ofElko.

[0030] The NSl Core module 104 operates on X(k) at each noise component sub- band k to generate corresponding noise suppression gains Gd_ns(k) for respective noise sub- bands. An embodiment of a noise adaptive algorithm is disclosed by US 6,035,048 to Diethorn. Another embodiment of a noise adaptive algorithm is disclosed by US 2003/0147538 Al to Elko.

[ 0031 ] Then, G_dns(k) is applied to the voice sub-band components X(k) at a gain combiner 108 in a single processing channel to generate a corresponding output Y(k). The noise suppression gains are combined at the gain combiner 108 prior to synthesis with the input signal in the single processing channel. In Fig. 1, the processing channel for the sub- bands X(k) is used for the single processing channel. Moreover, the single processing channel for the noise suppression gains Gd_ns(k) can be used as an alternative, single processing channel. In the single processing channel, synthesis filter bands of the SFB module 106 apply the noise suppression gains Gd_ns(k) to the voice sub-bands X(k) to attenuate the noise sub- bands present in the corresponding, single processing channel. The synthesis filter bands of the SFB module 106 are applied to Y(k) to synthesize the noise-suppressed sub-band components back to time-domain noise-attenuated signals Y(n). [0032] Fig. 1 discloses an example of a frequency sub band X(k) for spectral analysis processing. It should be understood that the quantity X(k) represents one or more frequency sub-bands of voice and noise of the input signal to be processed. The number of frequency sub-bands corresponds to the number of frequency bins for processing by the analysis filter bank AFB 102 and 112. For noise suppression, the number of sub-bands can be about ten in number or less or more, for example. The processing capacity is lessened by reducing the number of frequency bins for processing. However, the resolution of quality and trueness of the input signal demands a higher number of sub-bands for processing. The noise suppressor output for each frequency bin is obtained by applying the noise classification algorithm to correlate the sub-band noise waveforms within respective frequency bins, and then applying the noise suppression algorithm to calculate an estimated or predicted, noise suppression gain of each noise frequency bin. According to an embodiment of the invention, the suppression gain is calculated on the central frequency for each frequency bin. Alternatively, the suppression gain is calculated on a range of frequencies within the sub-band of frequencies for the frequency bin. [0033] Fig. 4 discloses another embodiment of a noise suppressor system comprised of an exemplary processor module 400, similar to the exemplary processor module 100, by having the AFB module 102, the NSl core module 104, the SFB module 106 and the gain combiner 118. Further, Fig. 4 discloses another embodiment of an exemplary processor module 410, similar to the exemplary processor module 110, by having the LMS based Calibration module 116, the AFB module 112 and the NS2 Core module 114. [0034] Prior to the invention, the output GN_Si2(k) of the NS2 Core module 114 was required to be applied to the sub-band components Ys(n) and Yd(n) of the AFB module 112 to generate noise suppression gain outputs, respectively, that suppress the noise sub-band components in each channel. Such a processing operation would be similar to that of the NSl core module 104 applying the output GN_Sii(k) to the output S(k) of the AFB module 102 to generate noise suppression gains that suppresses the noise sub-band components in the corresponding channel of the processor module 100. Thus, prior to the invention, two processing operations were required, either as two tandem stacked processing channels to remove noise or two parallel processing channels to remove noise. According to the invention, a gain combiner 118 is frequency tuned to the outputs of the frequency domain noise suppression gains. At the gain combiner 118, the output GN_Si2(k) of the core NS2 module 114 is applied to the output X(k) of the AFB module 102, wherein the output X(k) includes the sub-band components of the input signal X(n) passed by the AFB module 102. The gain combiner 118 is programmable for gain multiplication, addition, subtraction, or combination thereof. Accordingly, the invention combines frequency based processing of frequency bin sub-bands (especially noise) to calculate noise suppression gains obtained by applying two different algorithms in the single processing channel of the first exemplary noise suppressor processor module 100, and using the single processing channel of the processor module 100. This advantage is achieved because frequency-based processing is performed independently of phase, without requiring a common phase to combine the calculated noise suppression gains and the frequency bin sub-bands for noise-attenuation of the voice signal. [0035] Prior to the invention, a first processing channel for the processor module 110 was required to calculate suppression gain and perform attenuation of noise, such as, directional noise, for example. A second processing channel of the processor module 100 was required to calculate suppression gain and perform attenuation of noise, such as, fixed point noise (stationary noise) or dynamic noise (NSl), for example. These processing channels, when constructed as stacked processing channels, required stacked architectural processing capacity and copious computational load. According to the invention, savings in processing requirements are attained by eliminating the need for combining the calculated gains GN_Si2(k) with frequency bin sub-bands of the one processor module 110. Further savings in processing requirements are attained by providing the single SFB module 106 for synthesizing the time- domain signal Z(n) with the combined NSl and NS2 calculated gains, GNSii(k) and GNSi2(k), wherein the input Y(k) includes the combined inputs GN_Sii(k) and GN_Si2(k). The use of a shared synthesis filter bank of the SFB module 106 reduces computational load, reduces memory requirements and reduces processor capacity for processing two different algorithms in two parallel processing channels, or in stacked processing channels. Processing time is performed at a lower bit-rate in a shared synthesis filter bank as compared to stacked processing channels. Further, a true integration of two processing channels is attained. [0036] In converting back to the time-domain signal Z(n) at the SFB module 106, the

SFB module 106 applies a common phase for all n- frequencies of the output time-domain signal Z(n). It is noted that further savings in processing requirements are attained by eliminating the need to process the Ys(k) and Yd(k) frequency bin sub-bands back to the time-domain signals Xi(n) and S₂Oi). Thus, according to the invention, an input, noisy voice signal is processed in different processing channels by corresponding, different noise suppressor algorithms of any type to extract noise waveforms effectively. The noise waveforms are processed in the frequency domain, independent of phase. As a result, the two different processing channels can merge into a single processing channel for combining the extracted noise waveforms with suppression gains resulting from the application of two different algorithms, and for converting the noise-attenuated voice signal to the time-domain signal Z(n) for output in a single channel. [0037] Further, according to the invention, the noise suppression gains from NS2 and

NS 1 are combined in various ways including, but not limited to serial minimum gains or maximum gains, parallel minimum gains or maximum gains, multiply gains and combinations thereof. Fig. 4 discloses a gain combiner 418 in the form of a Gain Combination Operator which combines suppression gains produced by the exemplary NS2 noise suppressor algorithm with the sub-band waveforms X(k) produced by the exemplary NSl noise suppressor algorithm. In Fig. 4, the gain combiner 418 is downstream in series with the NSl Core module 104. By comparison, in Fig. 1, the gain combiner 118 is upstream in series with the NSl Core module 104. The Gain Combiner Operator 418, Fig. 4, is interchangeable with the gain combiner 118, Fig. 1, and is programmable with different settings enabling different solutions A, B or C to combine the gains of NSl and NS2 that will lead to different subjective qualities.

[0038] In Solution A, the gain combination operator is taken to be min { ., .} as follows: G(k) = min{G_dns (k), G_sns (k)} for all k (1) wherein, k is the frequency bin number, G(k) is the noise suppression gain of the NS2 module at frequency bin k, G(k) is the noise suppression gain of the NSl module at frequency bin k, G(k) is the gain combination output at frequency bin k, which will be applied to X(k), the sub-band component of mono channel signal x(n).

[0039] In Solution B, the gain combination operator is taken to be multiplier as follows:

G(k) = G_dm (k)x G_sns (k) for all k (2)

This solution is also equivalent to the application of NSl gain processing first and then NS2 gain processing shown in Figure 4. Although the NS2 processing in this solution and in Fig. 1 is exactly the same, note that the non-linearity is within the core NSl module and hence applying the NS2 processed results (i.e. suppression gain vector) to the point before or after the core NSl module would lead to the different results. As a result Solution B of Fig. 4 is not equivalent to Fig. 1. [0040] In Solution C, the gain combination operator is taken to be max { ., .} as follows:

G(k) = m_aχ{G_dns (k), G_sm (k)} for all k (3)

Solution C is more conservative than Solution A, and always will lead to the least amount of suppression. [0041] Fig. 5 discloses a (voice) signal transmission system 500. Two microphones

502, 504 (Mic 1, Mic 2), respectively, feed the mono-channel signal X(n) for processing by a stationary noise suppression algorithm or a NS 1 noise suppression algorithm, and feed the two-channel signals Xl(n) and X2(n) for processing by the NS2 noise suppression algorithm. A switch 506 selectively inputs the voice signals received by both microphones 502 and 504 for combination in a combiner 508. A noise suppressor system block 510 combines the modules 100 and 110 of Fig. 1, or the modules 400 and 410 of Fig. 4. In the embodiments of the system 500 that has a speaker 512 for output of a received audio signal V(n), the signal V(n) is protected from analog echo by being coupled through an Analog Echo Cancellation (AEC) module 514 to another combiner 516 of the system 500. [0042] Embodiments of the microphone inputs are suitable for a voice transmission system 500, including but not limited to (a) a cellular telephone appliance, (b) a land-line telephone appliance, (c) a VOIP telephone appliance, (d) a GPS telephone appliance, (e) a wireless microphone, (f) a wireless or plug-in headset or (g) a channel selective voice transceiver for one of, or both of, radio bandwidth and television bandwidth. [0043] According to the invention, a voice signal is processed with different noise suppressor algorithms to calculate noise suppression gains for different noise frequency bins, and the suppression gains are combined in a single processing channel to apply noise suppression on a single channel voice signal. Advantageously, the application of different algorithms perform better than a single algorithm, and the single processing channel reduces the processor capacity requirements in chip architecture to attenuate noise in a voice signal with different noise suppression algorithms. [0044] The invention combines multiple noise suppressors applying different algorithms. An embodiment of the invention will be described by referring to a dynamic noise suppressor (NSl) combined with a spatial noise suppressor (NS2). The invention is not limited to the NSl and NS2 embodiments, which are described herein by way of example, representing multiple noise suppressors applying different algorithms. The spatial noise suppressor NS2 is an adaptive noise suppressor added to two microphone inputs (two audio channel inputs), which provides better performance than one channel input (one microphone). Further, the invention is not limited to a noise suppressor added to two channel inputs, and embodiments of the invention includes noise suppressors added to one or more audio channel inputs to attenuate the level of background noise. [0045] This description of the exemplary embodiments is intended to be read in connection with the accompanying drawings, which are to be considered part of the entire written description. In the description, relative terms such as "lower," "upper," "horizontal," "vertical,", "above," "below," "up," "down," "top" and "bottom" as well as derivative thereof (e.g., "horizontally," "downwardly," "upwardly," etc.) should be construed to refer to the orientation as then described or as shown in the drawing under discussion. These relative terms are for convenience of description and do not require that the apparatus be constructed or operated in a particular orientation. Terms concerning attachments, coupling and the like, such as "connected" and "interconnected," refer to a relationship wherein structures are secured or attached to one another either directly or indirectly through intervening structures, as well as both movable or rigid attachments or relationships, unless expressly described otherwise.

[0046] Patents, patent applications and publications referred to herein are expressly incorporated in their entireties herein. Although the invention has been described in terms of exemplary embodiments, it is not limited thereto. Rather, the appended claims should be construed broadly, to include other variants and embodiments of the invention, which may be made by those skilled in the art without departing from the scope and range of equivalents of the invention.

Claims

What is claimed is:

1. A method comprising the steps of: (a) processing an input electrical signal representing a sound with a first noise suppressor algorithm to provide first processed electrical signals; (b) processing the input electrical signal with a second noise suppressor algorithm to provide second processed electrical signals; (c) combining the processed first electrical signals and processed second electrical signals in a single processing channel together with frequency sub-bands of the input electrical signal; (d) processing the combined processed first and second electrical signals in the single processing channel together with frequency sub-bands of the input electrical signal to form a time-domain output signal having reduced noise relative to the input electrical signal; and (e) outputting the output signal.

2. The method of claim 1, comprising: processing the combined, processed first and second electrical signals with a synthesis filter bank to form the time-domain output signal having reduced noise relative to the input electrical signal..

3. The method of claim 1, further comprising: combining the processed first electrical signals comprising noise suppression gains and the processed second electrical signals comprising second noise suppression gain thereof; and applying the noise suppression gains to frequency sub-bands of the input signal to form the time-domain output signal having reduced noise.

4. The method of claim 1, comprising: combining the processed first electrical signals and processed second electrical signals in a single processing channel, wherein the processed first electrical signals comprise first suppression gains for attenuating gains in noise frequency bins, and the processed second electrical signals comprise second suppression gains for attenuating gains in noise frequency bins; applying the combined first and second suppression gains to frequency sub-bands of the input electrical signal; and converting the sub-bands of the input signal to said time-domain output signal.

5. The method of claim 1, comprising: inputting the input electrical signal for processing by the first noise suppressor algorithm with a single first microphone; and inputting the input electrical signal for processing by the second noise suppressor algorithm with the first microphone and a second microphone, respectively.

6. A noise suppression system comprising: a first noise suppressor algorithm that processes an input electrical signal representing a sound to provide first processed electrical signals; a second noise suppressor algorithm that processes the input electrical signal to provide second processed electrical signals; a combiner combining the first and second processed electrical signals with frequency sub-bands of the input electrical signal to form a combined signal; and a processing module that processes the first and second processed electrical signals with the frequency sub-bands of the input electrical signal to form a time-domain output signal of reduced noise relative to the input electrical signal.

7. The system of claim 6, comprising: a single processing channel including the combiner and the processing module, wherein the processing module comprises a synthesis filter bank that processes the combined, first and second processed electrical signals with said frequency sub-bands of the input electrical signal to form the output signal.

8. The system of claim 6, comprising: first and second analysis filter banks that process the input electrical signal to form first and second noise frequency bins, respectively, for processing by the first and second noise suppressor algorithms, respectively.

9. The system of claim 6, wherein the first and second processed electrical signals comprise first and second noise suppression gains calculated by the first and second noise suppression algorithms, respectively; and the combiner combines the first and second noise suppression gains in a single processing channel.

10. The system of claim 6, further comprising: a voice transmission system, including but not limited to (a) a cellular telephone appliance, (b) a land-line telephone appliance, (c) a VOIP telephone appliance, (d) a GPS telephone appliance, (e) a wireless microphone, (f) a wireless or plug-in headset or (g) a channel selective voice transceiver for one of, or both of, radio bandwidth and television bandwidth.

1/3