JP2003526109A - Channel gain correction system and noise reduction method in voice communication - Google Patents

Abstract

(57) Abstract: A system and means for noise suppression of an audio processing system (108) are presented. A gain estimator (220) determines a gain for each frame of the input signal, and thereby determines a noise suppression level. If there is no speech in the frame, the gain is set to a predetermined minimum value. When voice is present in the frame, the gain adjuster (224) determines a gain coefficient for each predetermined one frequency channel set. The gain factor for each channel is a function of the SNR (signal-to-noise ratio) of the voice for that channel. The SNR estimator (210b) generates a channel SNR based on the channel energy estimate provided by the energy estimator (206b) and the channel noise energy estimate provided by the noise energy estimator (214b). The noise energy estimator (214b) updates the estimated value when the speech detector (208) determines that there is no speech in the frame.

Description

Detailed Description of the Invention

TECHNICAL FIELD The present invention relates to voice processing. More particularly, the present invention relates to noise suppression systems and methods used in speech processing.

BACKGROUND ART The transmission of voice by digital techniques is especially useful for cellular telephones and personal communication systems (P
It has been widely used in application fields such as CS). This also generated interest in improving voice processing techniques. One area where improvements have been made is the development of noise suppression techniques.

Noise suppression in speech communication systems generally serves the purpose of improving the overall quality of a desired audio signal by filtering environmental background noise from the desired speech signal. This voice enhancement process is especially necessary in environments with unusually high levels of ambient background noise, such as airplanes, moving vehicles, and noisy factories.

One noise suppression technique is spectral subtraction, a technique that modifies the gain of the spectrum. Using this scheme, the input audio signal is divided into multiple frequency channels, which causes a particular frequency channel to be attenuated according to its noise energy content. Utilizing the background noise estimate for each frequency channel,
The signal-to-noise ratio (SNR) of the voice on that channel is generated and this SNR ratio is used to calculate the gain factor for each channel. Next, the amount of attenuation of a specific channel is determined by this gain coefficient. The attenuated channels are recombined to produce a noise suppressed output signal.

In special applications with relatively high background noise environments, most noise suppression techniques exhibit significant performance limits. One example of such an application is the vehicle speakerphone option for cellular mobile communication systems. This speakerphone option allows the driver of the vehicle to operate hands-free. Hands-free microphones are typically mounted above a visor, and placed at a considerable distance from the user. This remote microphone has a land-end due to noise conditions such as roads and wind.
A bad SNR is provided to the side. Speech received at the land end is usually intelligible, but continuous exposure to such background noise levels often increases listener fatigue.

For the noise suppression system to work properly, it is important to accurately determine the SNR of the voice. However, due to the limitations of currently available noise detectors,
Accurately determining the SNR of an audio signal is difficult. The spectral subtraction technique updates the background noise estimate during periods of absence of speech. In the absence of speech, the measured spectral energy is due to noise, so the noise estimate is updated based on the measured spectral energy. Therefore, it is important to distinguish between the presence period and the absence period of speech to obtain accurate noise energy for calculating SNR.

One example technique for voice detection uses a voice metric calculator to perform noise update value decisions. Speech metric is a measure of the overall speech-like characteristics of channel energy. First, a raw SNR estimate is used to determine a voice metric table, which yields a voice metric value for each channel. The individual channel voice metric values are summed into an energy parameter, which is compared to the background noise update threshold. If the total voice metric is greater than or equal to this threshold, the signal is said to contain voice. If the total voice metric is less than the threshold, the input frame is considered noise and
Background noise update is performed. However, in the case of high background noise, sudden background noise, increasing noise sources, etc., the SNR measurement will be large, resulting in a high voice metric value, which invalidates the noise estimate update.

A refinement of the speech metric calculator technique measures the deviation of the channel energy. In this method, noise is assumed to exhibit constant spectral energy over time, while speech exhibits variable spectral energy over time. Therefore, the channel energies are integrated over time, so that the speech is detected when the deviation of the channel energies is quite large, while the noise is detected when there is little deviation of the channel energies. A speech detector that measures the channel energy deviation detects a sudden change in noise level. However, the channel energy deviation method gives inaccurate results when the input speech signal is a constant energy signal. Furthermore, in the case of a gradually increasing noise source, when the input energy changes, the energy deviation increases,
Therefore, even if the noise estimation update value is necessary, it becomes invalid.

In addition to accurate speech detectors, noise suppression systems must properly adjust channel gain. The channel gain should be adjusted so that noise is suppressed without sacrificing voice quality. One way to adjust the channel gain is to calculate the gain as a function of the overall noise estimate and the SNR of the speech signal. In general, as the overall noise estimate increases, the gain factor for a given SNR decreases. A low gain coefficient indicates a high attenuation coefficient. This technique imposes a minimum gain value to prevent channel gain over-attenuation when the overall noise estimate is very high. By using a minimum gain value that is strongly clamped, a tradeoff between noise suppression and voice quality is derived. When the clamp is relatively low, noise suppression is improved but voice quality is degraded. If the clamp is relatively high, the noise suppression is degraded but the voice quality is improved.

In order to provide an improved noise suppression system, it is necessary to point out the limitations of current techniques for voice detection and channel gain calculation. These problems and deficiencies are solved by the present invention as described below.

DISCLOSURE OF THE INVENTION The present invention is a system and method for noise suppression used in a speech processing system. It is an object of the invention to provide a voice detector which determines the presence of voice in the input signal. Reliable speech detectors are needed to determine the signal-to-noise ratio (SNR) of speech. If the speech is determined to be absent, the input signal is designated as a noise signal in its entirety and the noise energy is measured. Next, the noise energy is used to determine the SNR. Another object of the invention is to provide an improved gain measurement element for suppressing noise.

According to the invention, the noise suppression system comprises a speech detector which determines whether speech is present in the frame of the input signal. The presence / absence of voice is determined based on the SNR scale of the voice in the input signal. The SNR estimator estimates the SNR based on the signal energy estimation value generated by the energy estimator and the noise energy estimation value generated by the noise energy estimator. The voice presence determination is also based on the coding rate of the input signal. In a variable speed communication system, each input frame has a code acceleration (encor
ding rate). In general, this speed is
The frames that contain speech are assigned a high speed, while the frames that do not contain a speech are assigned a low speed, because they depend on the level of (ity). In addition, the voice presence determination may be based on one or more modal measures that describe the characteristics of the input signal. If it is determined that the voice does not exist in the input frame,
The noise energy estimator updates the noise energy estimate.

The channel gain estimator determines the gain for a frame of the input signal. If no speech is present in the frame, the gain is set to a certain minimum value. If present, the gain is determined based on the frequency content of the frame. In a preferred embodiment, the gain factor is determined for each of the predefined set of frequency channels. For each channel, it is determined according to the SNR of the voice on that channel. It is defined for each channel using a function suitable for the characteristics of the frequency band in which the channel exists. Generally, for a predefined frequency band, the gain is set to increase linearly with increasing SNR. In addition, the minimum gain for each frequency band may be adjustable based on environmental characteristics. For example, a user selectable minimum gain may be realized. The channel SNR is based on the channel energy estimate generated by the energy estimator and the channel energy estimate generated by the noise energy estimator. Using the gain factor,
Adjusting the gain of the signals on the various channels, the gain adjusted channels are combined to produce a noise suppressed output signal.

BEST MODE FOR CARRYING OUT THE INVENTION The features, objects and advantages of the present invention will be apparent from the detailed description given below with reference to the drawings in which like reference numerals indicate like elements throughout. Will.

In voice communication systems, noise suppressors are typically used to suppress unwanted environmental background noise. Most noise suppressors operate to estimate the background noise features of the input data signal in one or more frequency bands and subtract the average of this estimate from the input signal. The average background noise estimate is updated during periods of absence of speech. The noise suppressor needs to accurately determine the background noise level in order to operate properly. In addition, the noise suppression level must be adjusted correctly based on the speech and noise characteristics of the input signal. These requirements are handled by the noise suppression system of the present invention.

An exemplary speech processing system 100 in which the present invention is implemented is shown in FIG. The system 100 includes a microphone 102, an A / D converter 104, a voice processor 106, a transmitter 110, and an antenna 102. The microphone 102 may be located in a cellular telephone with the other elements shown in FIG. Alternatively, the microphone 102 may be a hands-free microphone, which is a vehicle speakerphone option for cellular communication systems. The assembly of a vehicle speakerphone is sometimes called a car kit. The noise suppression function is especially important when the microphone 102 is part of a car kit. Since hands-free microphones are typically located some distance from the user, received acoustic signals tend to have poor SNR due to road and wind conditions.

With continued reference to FIG. 1, an input audio signal containing voice and / or background noise is received by microphone 102. The input audio signal is converted by the microphone 102 into an electroacoustic signal represented by the item s (t). This electroacoustic signal may be converted from analog signals to pulse code modulation (PCM) samples by A / D converter 104. In one exemplary embodiment, PCM samples are output from A / D converter 104 at a rate of 64 kbps and are represented as signal s (n) as shown in FIG. The digital signal s (n) is received by the voice processor 106 which comprises a noise suppressor 108 along with other elements. Noise suppressor 108
Suppresses noise in the signal s (n) according to the invention. In carkit applications, noise suppressor 108 measures the level of background ambient noise and adjusts the gain of the signal to mitigate the effects of such ambient noise. In addition to the noise suppressor 108, the speech processor 106 typically comprises a voice coder, or vocoder (not shown), which extracts parameters associated with the model of human speech production. To compress the sound. The voice processor 106 also includes an echo canceller (not shown), which eliminates acoustic echo due to feedback between the speaker (not shown) and the microphone 102.

Following the processing by voice processor 106, the signal is output to transmitter 110, which may employ code division multiple access (CDMA), time division multiple access (TDMA), or frequency division. Modulation is performed according to a predetermined method such as a multiple access method (FDMA). In the exemplary embodiment, transmitter 110 is assigned to the assignee of the present invention and is incorporated herein by reference, "Spread Spectrum Multiple Access Communication System Using Satellite or Terrestrial Repeater" (SPREAD SPEC).
TRUM MUTPLE ACCESS COMMUNICATION SY
STEM USING SATELLITE OR TERRESTRIAL
The signal is modulated according to the CDMA format as described in U.S. Pat. No. 4,901,307 entitled "REPEATERS". Then, the transmitter 110 up-converts and amplifies the modulated signal, and the modulated signal is transmitted from the antenna 112.

It should be appreciated that the noise suppressor 108 may be implemented as a voice processing system different from the system 100 of FIG. For example, the noise suppressor 108 may be utilized in email applications with voice and options. In such an application, transmitter 110 and antenna 112 of FIG. 1 are not needed. Instead, the noise suppressed signal is formatted by the voice processor 106 and sent over the email network.

An exemplary embodiment of the noise suppressor 108 is shown in FIG. The input audio signal is received by the pre-processor 202 as shown in FIG. Advance processor 20
2 creates the input signal to be noise suppressed by performing preemphasis and frame generation. Pre-equalization redistributes the output spectral density of a speech signal by emphasizing the high frequency speech components of the signal. By performing the wide-area filtering function, the pre-equalization process is performed in the frequency domain
Improve the SNR of these components in the (domain). The pre-processor 202 also generates frames from samples of the input signal. In a preferred embodiment, a 10 ms frame of 80 samples / frame is generated. These frames may overlap the samples to improve processing accuracy. These frames may be generated by windowing and zero padding the samples of the input signal. Preprocessed (prepro
The cessed) signal is output to the conversion element 204. In a preferred embodiment, transform element 204 produces a 128-point fast Fourier transform (FFT) for each frame of the input signal. However, it should be understood that alternative schemes may be used to analyze the frequency components of the input signal.

These converted components are supplied to the channel energy estimator 206a, where the N channels of converted signals are used to generate energy estimates for each channel. For each channel, one technique for updating the channel energy estimates an update value that is the current channel energy smoothed with respect to the channel energy of the previous frame as follows: E _u ( t) = αE _ch + (1−α) E _u (t-1) (1) Here, the updated estimated value E _u (t) is the current channel energy E _ch and the previous estimated channel noise energy E _u ( t-1) and is defined as a function of

In a preferred embodiment, the energy estimate for the low frequency channel and the energy estimate for the high frequency channel are determined such that N = 2. The low frequency channel corresponds to the frequency range of 250-2250 Hz, while the high frequency channel corresponds to the frequency range of 2250-3500 Hz. The current channel energy of the low frequency channel is determined by summing the energy of the FFT points corresponding to 250-2250 Hz, and the current channel energy of the high frequency channel is summing the energy of the FFT points corresponding to 2250-3500 Hz. May be determined by

These energy estimates are provided to speech detector 208, which determines whether speech is present in the received audio signal. SN of voice detector 208
R estimator 210a determines the signal-to-noise ratio (SNR) of the speech on each of the N channels based on both the channel energy estimate and the channel noise energy estimate. The channel noise energy estimate is provided by the noise energy estimator 214a, but generally corresponds to the smoothed noise energy estimate on the previous frame that does not contain speech.

The voice detector 208 also comprises a rate determining element 212, which selects the data rate of the input signal from a predetermined set of data rates. In some communication systems, the data is encoded such that the associated data rate changes from frame to frame. This is known as a variable rate communication system. A voice coder that encodes data based on a variable rate scheme is commonly referred to as a variable rate vocoder. An example embodiment of a variable rate vocoder, "Variable Rate Vocoder" (VARI), assigned to the assignee of the present invention and incorporated herein by reference.
US Pat. No. 5,414,7 entitled ABLE RATE VOCODER)
No. 96. A variable rate communication channel can be used to eliminate unnecessary transmission content when there is useless audio to transmit. In order to change the number of information bits in each frame as the voice activity changes,
The algorithm is used within the vocoder. For example, a vocoder with a set of four rates may have 16, 40, 80 or 17 depending on speaker activity.
Generate a 20 ms data frame containing one information bit. It is preferable to transmit each data frame within a fixed time by changing the transmission rate during communication.

Since the rate of the frame depends on the voice activity during the time frame, determining the rate will provide information as to whether voice is present. In systems utilizing variable rates, the decision that a frame should be encoded at the highest rate generally indicates the presence of speech, while the decision that a frame should be encoded at the lowest rate generally Indicates absence. Intermediate rates generally indicate transitions between the presence and absence of speech.

Rate determining element 212 may be implemented in any of a plurality of rate determining algorithms. Such a rate determination algorithm, once assigned to the assignee of the present invention and incorporated herein by reference, is a "method and apparatus for reduced rate variable rate vocoding" (METHOD AND APPARATG.
US FOR PERFORMING REDUCED RATE VARIA
Co-pending US Patent Application No. 0 entitled BLE RATE VOCODING)
No. 8 / 286,842. This technique uses mode measures
), Which provides a set of rate decision criteria. The first modal measure is the Target Matched Signal-to-Noise Ratio (TMSNR) based on the previous coded frame, which shows how well the coding model is by comparing the synthesized speech signal with the input speech signal. It is information about what is being done. The second mode measure is the normalized autocorrelation function (NACF), which measures the periodicity of speech frames. The third modal measure is the zero crossings (ZC) parameter, which measures high frequency components in the input speech frame. The fourth measure, the predictive gain derivative
The n gain differential (PGD) is for determining whether the encoder maintains its prediction efficiency. The fifth measure is energy differential.
) (ED), which compares the energy in the current frame with the average frame energy. Using these modal measures, the rate determination logic selects the coding rate of the incoming frame.

Although rate determining element 212 is shown in FIG. 2 as an element included in noise suppressor 108, it is instead intended to provide rate information to noise processor 108 by another component of speech processor 106. It should be understood that it may be (Figure 1). For example, audio processor 106 may include a variable rate vocoder (not shown) that determines the coding rate for each frame of the input signal. Instead of having the noise suppressor 108 determine the rate alone, the rate information may be provided to the noise suppressor 108 by a variable rate vocoder.

It should also be appreciated that instead of determining the rate and determining the presence of speech, the speech detector 208 may use a subset of modal measures that contribute to the rate determination. For example, a NACF element (not shown) may be used in place of the rate determining element 212, which, as already mentioned, measures the periodicity of the speech frame. NACF is evaluated according to the following relationship:

[Equation 1]

Where N is the number of samples in the speech frame and t ₁ and t ₂ are the boundaries within the T samples that are the targets for evaluating the NACF. NACF is a formant
(formant) residual signal e (n). The formant frequency is the resonance frequency of voice. The voice signal is filtered using a short-term filter to obtain the formant frequency. The residual signal obtained after filtering by this short term filter is the formant residual signal, which contains the long-term speech information of the signal, such as the pitch.

The NACF mode measure is suitable for determining the presence of speech, since the periodicity of the signal containing vocalized speech differs from the signal not containing vocalized speech. Spoken speech tends to be characterized by periodic components. In the absence of vocalized speech, the signal generally has no periodic component. Therefore, the NACF measure may be a good indicator used by the voice detector 208.

Speech detector 208 may use a measure such as NACF in place of rate determination results in situations where it is not practical to generate rate determination results. For example, if the rate determination result is not available from the variable rate vocoder and the noise processor 1
If the 08 does not have the processing power to generate its own rate determination result, NAC
Modal measures such as F provide the desired alternative. This is the case, for example, in car kit applications where processing power is generally limited.

In addition, it should be appreciated that the voice detector 208 makes a decision regarding the presence of voice based solely on the rate decision result, the mode metric, and the SNR estimate. Although any further measures should improve the accuracy of the decision, any one of these measures alone may give reasonable results.

The rate determination result (or mode measure) and the SNR estimate generated by SNR estimator 210 a are provided to speech decision element 216. The voice decision element 216 makes a decision based on the input whether voice is present in the input signal. The decision regarding the presence of speech determines whether or not to update the noise energy estimate. The noise energy estimate is used by the SNR estimator 210a to determine the SNR of the speech in the input signal. This SNR is then used to calculate the level of attenuation of the input signal for noise suppression. If it is determined that speech is present, speech determination element 216 opens switch 218a to prevent noise estimator 214a from updating the noise energy estimate. If it is determined that no speech is present, then the input signal is presumed to be noise and the speech decision element 216 closes switch 218a causing the noise energy estimator 218a to update the noise estimate. Although shown as switch 218a in FIG. 2, it should be understood that the enable signal provided to the noise energy estimator 214a from the voice decision element 216 performs the same function.

In one preferred embodiment where two channel SNRs are evaluated, the voice decision element 216 determines the noise update decision based on the following procedure.
) Is generated: If (rate = minimum value) ((chsnr1> T1) or (chsne2> T2)), if (rate count (ratecount)> T3), the noise estimation value is updated. Otherwise rate count ++ otherwise update noise estimate rate count = 0 otherwise rate count = 0 channel SNR estimate provided by SNR estimator 210a is chsnr1 and ch
Represented by snr2. The rate of the input signal provided by the rate determining element 212 is represented by rate. A counter, rate count, keeps track of the number of frames based on certain conditions described below.

The voice judgment element 216 determines that the rate is the minimum value of the variable rates, and chsn
If r1 is greater than the threshold T1 or chsnr2 is greater than the threshold T2 and the rate count is greater than the threshold T3, then there is no voice and the noise estimation ch
Judge that the same should be updated. Rate is the minimum value, chsnr1 is T1
Greater than or chsnr2 is greater than T2 but the rate count is less than T3, the rate count is incremented by one but the noise estimate is not updated. A counter, a rate count, counts the number of frames with the lowest rate but has high energy in at least one of the channels at the same time, so that the noise level suddenly increases or the noise source gradually increases. Detect the case. The counter, which is an indicator that the high SNR signal does not contain voice, is set to count until voice is detected in the signal. One preferred embodiment is T1 = T2 = 5 dB where a 10 ms frame is evaluated.
, T2 = 100 frames are set.

If the rate is at a minimum value, chsnr1 is less than T1 and chsnr2 is less than T2, the speech decision element 216 decides that there is no speech and therefore the noise estimate should be updated. To do. In addition, the rate count is reset to zero.

If the rate is not at the minimum value, the voice decision element 216 determines that the frame contains voice and therefore the noise estimate should not be updated and the rate count is reset to zero. To be done.

Instead of using a rate measure to determine the presence of speech, NACF
It should be remembered that mode measures such as measures can be used. The speech decision element 216 may utilize the NACF measure to determine the presence of speech, so the noise update decision is performed according to the following procedure: If ((pitch Present == false (FALSE) ((chsnr1> TH1) or (chsnr2> TH2) if (pitchCount> TH3) update the noise estimate otherwise pitchCount + + otherwise update the noise estimate pitchCount = 0 Otherwise pitchCount = 0 where pitchPresent is defined as follows: pitchPresent = TRUE NACF Nucount = 0 if (NACF> TT1) otherwise (TT2 ≦ NACF ≦ TT1) If (NACFCOUNT> TT3) pitchPresent = truth otherwise pitchPresent = false NACFCOUNT + + otherwise pitchPresent = false NACFCOUNT = 0 Furthermore, SNR Channel SNR estimates Joki 210a is supplied chsnr1 and ch
It is represented by snr2. The NACF element (not shown) produces pitchPresent, which is a measure of the presence of the above pitch. The counter, pitchCount, keeps track of the number of frames based on certain conditions described below.

The scale pitchPresent determines that pitch exists when NACF is greater than the threshold value TT1. If the NACF is in the intermediate range (TT2 ≦ NCF ≦ TT1) for a plurality of frames larger than the threshold value TT3, it is determined that the pitch exists. The counter, NACFcount,

[Equation 2]

Keep track of the number of frames where In a preferred embodiment, 10 ms frames are evaluated with TT1 = 0.6, TT2 = 0.4, TT3 = 8 frames.

The speech decision element 216 indicates that the pitchPresent measure indicates that there is no pitch (pitchPtrsent = false) and chsnr1 is greater than the threshold TH1 or
If chsnr2 is greater than the threshold value TT2 and pitchCount is greater than the threshold value TH3, it is determined that there is no voice and therefore the noise estimate should be updated. pitchPresent = false and chsnr1 is greater than TH1 or chnsr2 is T
If it is greater than H2 but pitchCount is less than TH3, pitchCount is incremented by one but the noise estimate is not updated. The counter pitchCount is used to detect sudden increases in noise level and gradual increase in noise sources. In one preferred embodiment, 1
The conditions of T1 = T2 = 5 dB and T2 = 100 frames in which 0 ms frame is evaluated are set.

If pitchPresent indicates that there is no pitch and chsnr1 is less than TH1 or chsnr2 is less than TH2, the speech decision element 216 has no speech and therefore should update the noise estimate. It is determined that In addition, pitchC
opunt is reset to zero.

If pitchPresent indicates that a pitch is present (pitchPresent = truth), then the speech decision element 216 decides that the frame contains speech and therefore the noise estimate should not be updated. However, pitchCount is reset to zero.

If it is determined that no speech is present, switch 218a is closed and noise energy estimator 214a updates the noise estimate. Noise energy estimator 214a
Generally produces a noise energy estimate for each of the N channels of input signal. Since there is no voice, it is assumed that all the energy is due to noise. For each channel, the noise energy update is estimated to be the current channel energy smoothed with respect to the channel energy of the previous frame that does not contain speech. For example, the updated estimate is obtained based on the following relation: E _n (t) = βE _ch + (1-β) E _n (t-1), (3) where the updated estimate The value E _n (t) is defined as a function of the current channel energy E _ch and the previous estimated channel noise energy E _n (t-1). In one exemplary embodiment, β = 0.1 is set. The updated channel noise energy estimate is the SNR
It is provided to the estimator 210a. These channel noise energy estimates are used to obtain the channel SNR estimate update for the next frame of the input signal.

The decision regarding the presence of speech is also provided to the channel gain estimator 220. The channel gain estimator 220 determines the gain and thus the noise suppression level for the frame of the input signal. If the voice decision component 216 determines the presence of that voice,
The gain for the frame is set to a predetermined minimum gain level. Otherwise,
Gain is determined as a function of frequency. In the preferred embodiment, the gain is calculated based on the graph shown in FIG. Although shown graphically in FIG. 3, it should be understood that the functions shown in FIG. 3 may be implemented as a look-up table in channel gain estimator 220.

It can be seen in FIG. 3 that the preferred embodiment of the present invention defines a separate gain curve for each L frequency band. In FIG. 3, three bands (L = 3
) Is displayed, L may be any number of 1 or more. Thus, the gain factor for the low band channel is determined using the low band curve, the gain factor for the intermediate band channel is determined using the intermediate band curve, and the gain factor for the high band channel is determined. May be determined using a high band curve.

Noise suppression may be implemented using only one gain curve (L = 1) for the input signal, but when multiple bands are used, the voice quality is less likely to deteriorate. Have been found. In the case of environmental noise, such as road and wind noise, the energy of the noise signal is large at the lower frequencies and generally this energy decreases as the frequency increases.

In FIG. 3, a linear equation with a fixed slope and y-intercept is used to determine the gain factor for each band. The determination of the gain factor can be explained by the following relationship: gain [low band] (dB) = slope 1 * SNR + low band y-intercept; (4) gain [middle band] (dB) = slope 2 * SNR + (5) Gain [High Band] (dB) = Slope 3 * SNR + High Band y-Intercept; (6) The preferred embodiment specifies the low band as 125-375 Hz and the intermediate band as 3.
Designate as 75-2625 Hz and high band as 2625-4000 Hz.
The slope and y-intercept are determined empirically. Although the preferred embodiment uses the same slope 0.39 for each of the three bands, different slopes may be used for each frequency band. Furthermore, the low band y-intercept is -17d.
B, the middle band y-intercept is set to -13 dB, and the high band y-intercept is set to -13 dB.

In order to select the desired y-intercept, any feature will provide the user of the device with a noise suppressor. Thus, more noise suppression (lower y-intercept) may be chosen at the expense of speech degradation. Or y-
The intercept may be variable as a function of some unit of measure determined by the noise suppressor 108. For example, more noise suppression (lower y-intercept) may be desirable if excessive noise energy is detected during a given time period. Alternatively, less noise suppression (higher y-intercept) may be desirable if conditions such as bubbles are detected. During the bubble state, a background speaker is present and less noise suppression may be justified to prevent cutout of the main speaker. Another optional feature would provide for a selectable slope of the gain curve. Further, it may be found that curves other than the straight lines described by equations (4)-(6) are more suitable for determining the gain factor under certain circumstances.

For each frame containing speech, a gain factor is determined for each of the M frequency channels of the input signal, M being a predetermined number of channels to evaluate. The preferred embodiment evaluates 16 channels (M = 16). Referring again to FIG. 3, the gain factor for channels having frequency components within the low band is determined using the low band curve. The gain factor for channels with frequency components within the mid band is determined using the mid band curve. The gain factor for channels with frequency components in the high band range is determined using the high band curve.

For each channel evaluated, the channel SNR is used to derive the appropriate curve-based gain factor. In FIG. 2, the channel SNR is shown to be evaluated by the channel energy estimator 206b, the noise energy estimator 214b and the SNR estimator 210b. For each frame of the input signal, channel energy estimator 206b generates an energy estimate for each of the M channels of the transformed input signal, and the energy estimate is SNR estimator 210.
b. The channel energy estimate can be updated using the relationship in equation (1) above. If the speech decision component 216 determines that no speech is present in the input signal, the switch 218b is closed and the noise energy estimator 214b updates the channel noise energy estimate. For each of the M channels, the updated noise energy estimate is the channel energy estimator 2
06b based on the channel energy estimate. The updated estimate can be evaluated using the relationship in equation (3) above. The channel noise estimate is provided to SNR estimator 210b. Thus, the SNR estimator 210
b determines a channel SNR estimate for each voice frame based on the channel gain estimate for the particular voice frame, and a channel noise energy estimate is provided by noise energy estimator 214b.

Those skilled in the art will understand that the channel energy estimator 206a, the noise energy estimator 214a, the switch 218a, the SNR estimator 210a, the channel energy estimator 206b, the noise energy estimator 214b, and the switch 218b.
, And will each perform a similar function as the SNR estimator 210b.
Thus, although shown as separate processing components in FIG. 2, channel energy estimators 206a and 206b may be combined as one processing component and noise energy estimators 214a and 214b may be combined as one processing component. Switches 218a and 218b may be combined as one processing component, and SNR estimators 210a and 210b may be combined as one processing component. As a combined component, the channel energy estimator determines channel energy estimates for both the N channels used for speech detection and the M channels used for determining channel gain factors. Will do. Note that N = M is possible. Similarly, the noise energy estimator and S
The NR estimator will work for both N and M channels. The SNR estimator then provides N SNR estimates for the speech decision component 216 and M gain SNR estimates for the channel gain estimator 220.

The channel gain factors are provided by the channel gain estimator 220 to the gain adjuster 224. Gain adjuster 224 receives the FFT transformed input signal from transform component 204. The gain of the converted signal is adjusted appropriately according to the channel gain coefficient. For example, in the above embodiment where M = 16, the transformed (FFT) points belonging to a particular channel of the 16 channels are adjusted based on the appropriate channel gain factor.

The gain adjusted signal generated by the gain adjuster 224 is then converted to the conversion component 22.
6 is provided to invert, and in the preferred embodiment transform component 226 produces an inverse fast Fourier transform (IFFT) of the signal. If the input frame is formed of superimposed samples, post processing element 228 conditions the output signal for overlap. The post-process component 228 also performs deemphasis if the signal experiences preemphasis. De-emphasis attenuates the emphasized frequency components during pre-equalization. Pre-equalization /
The de-emphasis process effectively contributes to noise suppression by reducing noise components that are outside the range of processed frequency components.

The various processing blocks of the noise suppressor shown in FIG.
SP) or application specific integrated circuit (ASIC). Given the functionality of the present invention, one of ordinary skill in the art would be able to implement the present invention in a DSP or ASIC without undue experimentation.

Referring now to FIG. 4, a flowchart illustrating some of the steps involved in the process described in connection with FIGS. 2 and 3 is shown. Although shown as sequential steps, those skilled in the art will recognize that the order of some of the steps may be interchanged.

The process begins at step 402. In step 404, the transformed component 20
Reference numeral 4 converts the input audio signal into a converted signal, which is converted into an FFT signal. In step 406, the SNR estimator 210b determines the voice SNR for the M channels of the input signal based on the channel energy estimate provided by the channel energy estimator 206b and the channel noise energy estimate provided by the noise energy estimator 214b. To decide. In step 408, channel gain estimator 220 determines gain factors for the M channels of the input signal based on the frequencies of the channels. The channel gain estimator 220 sets the gain to the lowest level if it is found that there is no speech in the frame of the input signal.
Otherwise, the gain factor for each of the M channels is determined based on a predetermined function. For example, in FIG. 3, functions defined by linear equations with a fixed slope may be used, where each linear equation limits the gain for a given frequency band. In step 410, gain adjuster 224 uses the M gain factors to adjust the gains of the M channels of the converted signal. In step 412, the inverse transform component 226 transforms the gain adjusted transformed signal,
Produces a noise suppressed audio signal.

At step 414, the SNR estimator 210a uses the channel energy estimate provided by the channel energy estimator 206a and the channel noise energy estimate provided by the noise energy estimator 214a to determine N channels of the input signal. Determine the voice SNR for In step 416, the rate determining element 212 determines the coding rate for the input signal through analysis of the input signal. Alternatively, one or more mode means such as NACF may be determined. In step 418, the speech decision element 216 determines whether speech is present in the input signal based on the SNR provided by the SNR estimator 210a and the rate and / or mode means provided by the rate decision element. If at decision block 420 it is determined that no speech is present, then the input signal is assumed to be completely noisy, and at step 422 the noise energy estimator 214a.
Updates the noise estimate. Noise energy estimator 214a updates the noise estimate based on the channel energy determined by channel energy estimator 206a. Whether or not speech is detected, the procedure continues processing the next frame of the input signal.

The previous description of the preferred embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments without the use of inventive talents. . As such, the invention is not intended to be limited to the embodiments shown herein, but should be consistent with the broadest scope consistent with the principles and novel features disclosed herein.

[Brief description of drawings]

[Figure 1]   FIG. 4 is a block diagram of a communication system using a noise suppressor.

[Fig. 2]   1 is a block diagram showing a communication system according to the present invention.

FIG. 3 is a graph of frequency-based gain factors for implementing noise suppression according to the present invention.

4 is a flow chart illustrating an exemplary embodiment of the processing steps required for noise suppression as implemented by the processing element of FIG.

─────────────────────────────────────────────────── ─── Continued front page    (81) Designated countries EP (AT, BE, CH, DE, DK, ES, FI, FR, GB, GR, IE, IT, L U, MC, NL, PT, SE), OA (BF, BJ, CF) , CG, CI, CM, GA, GN, ML, MR, NE, SN, TD, TG), AP (GH, KE, LS, MW, S D, SZ, UG, ZW), EA (AM, AZ, BY, KG , KZ, MD, RU, TJ, TM), AL, AM, AT , AU, AZ, BA, BB, BG, BR, BY, CA, CH, CN, CU, CZ, DE, DK, EE, ES, F I, GB, GE, GH, HU, ID, IL, IS, JP , KE, KG, KP, KR, KZ, LC, LK, LR, LS, LT, LU, LV, MD, MG, MK, MN, M W, MX, NO, NZ, PL, PT, RO, RU, SD , SE, SG, SI, SK, SL, TJ, TM, TR, TT, UA, UG, UZ, VN, YU, ZW

Claims (35)

[Claims] 1. A noise suppressor for suppressing background noise of a speech signal, a signal-to-noise ratio (SNR) estimator for generating a channel SNR estimate for a predetermined first frequency channel set of the speech signal. A gain estimator for generating a gain coefficient based on the corresponding one of the channel SNR estimation values is provided for each frequency channel, and the gain coefficient is SN
And a gain adjuster that adjusts a gain level for each of the frequency channels based on the corresponding gain coefficient, the background noise of the audio signal being suppressed. Noise suppressor. 2. The noise suppressor of claim 1, wherein the gain function is frequency dependent. 3. The gain function is implemented as a look-up table.
Noise suppressor. 4. The noise suppressor according to claim 1, wherein the gain function is a linear function having a slope and a y-intercept. 5. The noise suppressor of claim 4, wherein a user can select the y-intercept. 6. The noise suppressor of claim 4, wherein the y-intercept can be adjusted based on measured noise characteristics of the audio signal. 7. The noise suppressor of claim 4, wherein the slope is user selectable. 8. The noise suppressor according to claim 4, wherein the slope can be adjusted based on a noise characteristic measurement value of the voice signal. 9. A voice detector for determining the presence of voice in the voice signal;
And a noise energy estimator for generating an updated channel noise energy estimate for each of the frequency channels when the speech detector determines that no speech is present in the speech signal, the updated channel The noise suppressor of claim 1, wherein a noise energy estimate is provided to the SNR estimator that produces the channel SNR estimate. 10. A signal-to-noise ratio (SNR) estimator, wherein the speech detector produces a channel SNR estimate for a predetermined second frequency channel set of the speech signal; and the second frequency. 10. The noise suppressor of claim 9, comprising: a voice determination element that determines the presence of voice according to the channel SNR estimate of a channel set. 11. The speech detector comprises a rate determination element for determining an encoding rate of the first variable rate set of the speech signal; the speech determination element further comprising the presence of speech according to the encoding rate. The noise suppressor according to claim 10, which determines. 12. The voice detector comprises a mode measuring element for determining at least one mode measure characterizing the voice signal; the voice determining element further comprising at least one mode measuring means 11. The noise suppressor of claim 10, which determines that is present. 13. The noise suppressor according to claim 12, wherein the mode measuring means is a normalized autocorrelation function (NACF). 14. Means for determining the presence of voice in the voice signal; Means for generating a channel signal-to-noise ratio (SNR) of one predetermined frequency channel set of the voice signal; Voice is present If the means for determining that there is a voice,
The gain function is defined for each frequency band set and each frequency band, and the gain coefficient is defined to increase with an increase in SNR. A gain coefficient is determined for each frequency channel based on a gain function of the frequency band including a channel, the channel gain coefficient being determined based on a gain function of a frequency band whose range includes the frequency channel; and A noise suppressor that suppresses background noise of the voice signal, comprising means for adjusting the gain level of each frequency based on the channel gain coefficient. 15. The means for determining a gain factor determines the minimum gain factor for each of the frequency channels when the means for determining the presence of voice determines that no voice is present. 14 noise suppressors. 16. The noise suppressor of claim 14, wherein the gain function is implemented as a look-up table. 17. The noise suppressor of claim 14, wherein the gain function is a linear function with slope and y-intercept. 18. The user can select the y-intercept.
7 noise suppressor. 19. The noise suppressor of claim 17, wherein the y-intercept strip is adjustable based on the measured noise characteristics of the audio signal. 20. The noise suppressor of claim 17, wherein the slope is user selectable. 21. The noise suppressor of claim 17, wherein the slope is not adjustable based on the measured noise characteristics of the audio signal. 22. Means for generating an updated channel noise energy estimate for each frequency channel when the means for determining the presence of speech determines that there is no speech in the speech signal, 15. The noise suppressor of claim 14, wherein the updated channel noise energy estimate is provided to means for generating an SNR estimate for updating the channel SNR estimate. 23. The means for determining the presence of voice   Means for determining the coding rate of one coding rate set of said speech signal; and   Means for determining the presence of speech according to the coding rate; 15. The noise suppressor of claim 14, comprising: 24. The means for determining the presence of speech has an SNR for a predetermined second set of frequency channels of the speech signal.
24. The noise suppressor of claim 23, comprising: means for generating an estimate; the means for determining the presence of speech further determines according to the SNR estimate. 25. Means for determining the presence of speech by means for determining at least one mode measuring means characterizing the speech signal;
And a means for determining the presence of voice according to the at least one mode measuring means. 26. The means for determining the presence of speech comprises: for another predetermined set of frequency channels of the speech signal,
26. The noise suppressor of claim 25, comprising means for generating an SNR estimate; said means for determining the presence of speech further determines according to said SNR estimate. 27. The noise suppressor of claim 25, wherein the mode measuring means is a normalized autocorrelation function (NACF). 28. Converting the audio signal and displaying at the frequency of the audio signal; Determining the presence of audio in the audio signal; Signal pair of a predetermined frequency channel set of the frequency. Noise ratio (SNR)
Generating a voice function in the voice signal, determining the gain coefficient for each frequency channel, defining a gain function for each frequency band set and for each frequency band, and A gain coefficient is defined to increase with an increase, so that a gain coefficient for each frequency channel is determined based on a gain function of the frequency band including the frequency channel; A means for suppressing background noise of an audio signal, comprising: adjusting a gain level for each channel; and inversely converting the frequency for which the gain is adjusted to generate an audio signal in which noise is suppressed. 29. The means of claim 28, wherein a minimum gain factor is determined for each of the frequency channels when it is determined that there is no voice in the voice signal. 30. The means of claim 28, wherein the gain function is a linear function with slope and y-intercept. 31. The step of determining the presence of speech produces an updated channel noise energy estimate for each of the frequency channels when determining that speech is not present in the speech signal, 29. The means of claim 28, wherein the updated channel noise energy estimate is used to generate the channel SNR estimate. 32. The step of determining whether speech is present comprises channel SNR of a predetermined second set of frequency channels of the speech signal.
Generating an estimate; and according to the channel SNR estimate of the set of the second frequency channels,
29. The means of claim 28, comprising the step of: determining whether voice is present. 33. The step of determining the presence of speech determines the coding rate of one variable rate set of the speech signal; and the step of determining the presence of speech according to the coding rate. 33. The means of claim 32, comprising: 34. The step of determining the presence of voice comprises the step of determining at least one mode measuring means characterizing the voice signal; and the presence of voice according to the at least one mode measuring means. 33. The means of claim 32, further comprising: determining. 35. The means of claim 34, wherein the mode measuring means is a normalized autocorrelation function (NACF).