JP2003517761A - Method and apparatus for suppressing acoustic background noise in a communication system - Google Patents

Abstract

(57) Abstract: A method and apparatus for suppressing acoustic background noise in a communication system. The operating signal to noise ratio (SNR) level is reliably evaluated by the SNR level estimator (295) from the channel energy (293) and background noise energy (294) values. The minimum gain factor and gain slope are adapted 290 according to the operating SNR level. Using these adapted values and the channel SNR, a channel gain is selected (233). If the channel SNR is below a certain threshold, the channel is completely noise-like, the gain factor selected is minimal, and the attenuation of the channel is maximal. If the channel SNR is quite large, the selected channel gain is 0 dB. If the channel SNR is an intermediate value, the gain factor selected is between the minimum and 0 dB.

Description

Detailed Description of the Invention

FIELD OF THE INVENTION The present invention relates generally to noise suppression, and more particularly to noise suppression in communication systems.

BACKGROUND OF THE INVENTION Noise suppression techniques in communication systems are well known. The goal of noise suppression techniques is to reduce the amount of background noise during speech coding so that the overall quality of the coded speech signal provided to the user is improved. Communication systems that perform voice coding include, but are not limited to, voice mail systems, portable radio systems, trunk line communication systems, aircraft communication systems, and the like.

As one of the noise suppression techniques, there is a spectrum subtraction method incorporated in a mobile radio system. In this technique, the voice input is divided into individual spectral bands (channels) by means of appropriate spectral dividers, and the individual spectral channels are attenuated based on the noise energy content of each channel. The spectral subtraction method uses an estimator of the background noise power spectral density,
The signal-to-noise ratio (channel SNR) of the voice in each channel is generated and then used to calculate the gain factor for each individual channel. The gain factor is then used as an input to modify the channel gain for each individual spectral channel. The channels are then recombined to produce a noise suppressed output waveform.
US Pat. No. 4,811,404 to Vilmur and US Pat. No. 5,659,622 to Ashley (both assigned to the assignee of the present application and both hereby). Each of which is incorporated herein by reference) discloses a method and apparatus for suppressing acoustic background noise in a communication system. An example of a spectral subtraction technique implemented in an analog mobile radio telephone system can be found in U.S. Pat. No. 4,811,404 by Burmer.

In the prior art represented by US Pat. No. 4,811,404, the minimum gain (ie maximum attenuation) coefficient applied to every channel is fixed to a constant value. Channels that are considered completely noise-like (low SNR values) are assigned this minimum gain factor and are maximally attenuated. On the other hand, a channel considered to be completely voice-like (high SNR value) is assigned a gain coefficient of 1.0 (0 dB) and is not attenuated at all. Channels that are somewhat voice-like (medium SNR values) are assigned gain factors between a minimum value and 1.0. The choice of minimum gain factor is
It is based on the following two conflicting requirements: That is, 1) the minimum gain coefficient is
The background noise should be sufficiently attenuated so that the noise-suppressed speech is more audible. Also, 2) the minimum gain factor must be large enough so that any suppression of unintentional but unavoidable weak speech sounds does not significantly degrade the speech intelligibility. The selected value for the normal minimum gain coefficient value is 0.2239 (-13 dB).

The above-described method using the fixed minimum gain coefficient value is effective when the ratio of the total audio signal power to the background noise power, that is, the SNR is relatively large, for example, 15d
When it is B or more, it functions to some extent. However, in recent years, the use of wireless telephones has become widespread, and the demand for the performance of the noise suppressor in acoustically harsh environments is ever increasing. For example, not only hands-free use in vehicles, but also places such as airfields and railway stations are rapidly becoming the normal operating environment for wireless telephones. The effect is that the normal expected signal-to-noise ratio (SNR) is poor and prior art noise suppression techniques are not adapted to these more severe operating environments. Therefore, there is a need for a more robust noise suppression system for communication systems that provides higher quality in these more harsh environments.

SUMMARY OF THE INVENTION The present invention provides a method for suppressing acoustic background noise in a communication system. The method comprises estimating a signal component and a noise component of the input signal to produce a signal to noise ratio estimate, and determining at least a maximum noise attenuation coefficient based on the signal to noise ratio estimate. Generating a gain function based on the maximum noise attenuation coefficient, and applying the gain function to the input signal to generate a noise suppressed signal for use in the communication system. Characterize.

Description of the Preferred Embodiments A noise suppression system implemented in a communication system has a low signal-to-noise ratio (S
NR) conditions provide improved quality levels, thereby extending the range of SNRs where noise suppression is useful. As shown in FIG. 2, the noise suppression system 109
Specifically adapts the minimum gain factor value based on the operating SNR level to the adaptive block 2
It incorporates 90. The minimum gain factor value increases as the SNR level decreases. This has the effect of reducing background noise attenuation, but does not suppress weak voice sounds that are difficult to detect in low SNR conditions, which allows higher clarity and improved overall quality. The operating SNR level 292, which serves as an input to the adaptation block 290, is reliably estimated by the SNR level estimator 295 from the values of the channel energy 293 and the background noise energy 294.

The channel gain calculator 233 calculates the gain coefficient applied to each channel as the channel S.
Calculate based on NR. It has two parameters: MIN_GAIN (
dB) and GAIN_SLOPE (dB / dB) are used. In FIG. 3, the operation of the channel gain calculator 233 can be described as follows. Channel SNR (d
If B) is less than some threshold (CH_SNR_THLD), ie the channel is completely noise-like, then the selected gain factor is the minimum or MI.
N_GAIN, the channel attenuation is maximum. On the other hand, if the channel SNR is fairly large, ie, the channel is nearly voice-like, then the selected channel gain is 0 dB (ie 1.0 on the linear scale) and this channel is not attenuated at all. . If the channel SNR is an intermediate value, that is, the channel is partially voice, then the selected gain factor will be between MIN_GAIN and 0 dB. The channel gain calculation (in dB) is represented by the following equation.

[0009]

[Equation 1]

[Equation 2]

[Equation 3] Typical values for the various parameters mentioned above used in the channel gain calculation are MIN_GAIN = -13 dB, GAIN_SLOPE = 0.39 dB, CH
_SNR_THLD = 2.25 dB.

The above method for calculating the channel gain coefficient is, for example, 15 dB when the ratio of the voice signal power to the background noise power (SNR) is sufficiently large.
In the above case, it works well. In this case, the distinction between speech and noise is clearer. Background noise is strongly attenuated (by MIN_GAINdB), strong voice sounds are not really attenuated, and weak voice sounds (mainly noisy channels) are attenuated slightly. Background noise is suppressed, and since there is no significant deterioration in speech recognition, speech quality is improved.

However, if the noise suppressor 109 is required to run at lower SNR levels, the above method for calculating the gain factor is inadequate. At low SNR levels, the distinction between speech and noise is ambiguous, especially for weak speech sounds. As a result, such speech is attenuated, resulting in loss of clarity. Background noise is attenuated, but the loss of clarity degrades the overall voice quality.

In order to improve noise suppressor performance at low SNR levels, the values of MIN_GAIN and GAIN_SLOPE are adjusted so that weak voice sounds are not attenuated as much as possible. For example, assume that the value of MIN_GAIN has increased from -13 dB to -10 dB. In this case, the background noise is less attenuated, but with less voiced sound and corresponding loss of clarity. This improves the overall voice quality. However, the modified parameter values are not optimal when the SNR level increases. Therefore, a possible solution to this is to reliably estimate the operating SNR level and adapt the values of the MIN_GAIN and GAIN_SLOPE parameters based on this estimated SNR level. With this solution, the noise suppressor 109 works satisfactorily over a wider range of SNR levels.

The specific contents of the solution proposed as described above will be described below. To estimate the SNR level, the voice energy and noise energy are estimated separately in dB and the difference between the two is taken. The speech energy in dB is a filtered version of the peak channel energy (dB) samples in the voiced frame. By requiring only the voiced frames and peak channel energy to be used, the effect of background noise on the voice energy estimation is minimized. The noise energy in units of dB is a filtered version of the total noise energy sample in the noise-only frame of all channels. Before using the channel energy as well as the noise energy for speech estimation,
The effect of the pre-emphasis filter on the channel energy is removed from the waveform table by using the pre-calculated coefficients. The waveform table coefficient is calculated from the magnitude of the inverse square spectrum in the pre-emphasis filter.

A list of various variables and parameters used to describe this example is as follows: μ1, μ2, δ, ε-filter coefficients. σ, Φ-Proportional constant. ch_eng (i) —A float variable that stores the channel energy, ie the average energy in the i th frequency channel.

First-A static boolean variable that is true only in the first frame. frame_count-A static integer variable indicating the frame number. update_flag-a boolean variable that forces the background noise energy estimate to be updated in preference to the update_flag. gain_slope-A float variable that stores the gain_slope value and is used to calculate the channel gain coefficient.

I—An integer variable used as a subscript. min_gain-A static float variable that stores the filtered value of the minimum gain value sample and is used to calculate the channel gain factor. min_gain_raw-A float variable that stores the minimum gain value sample calculated as a function of SNR level.

Noise_enrg_dB—A float variable that stores noise energy samples in units of dB. noise_enrg_dB_filt-A static float variable that stores the filtered version of the noise energy sample in units of dB.

Shape_table (i) —A float variable that stores the i th waveform table entry. snr-A float variable that stores the estimated SNR level in dB. spch_enrg_dB—A float variable that stores the voice energy sample in dB.

Spch_enrg_dB_filt—A static float variable that stores the filtered version of the speech energy sample in dB. update_flag-a boolean variable that indicates that the current frame is noise only and therefore the background noise energy estimate can be updated. vm_sum—An integer variable that stores the total audio measurements for different channels.

GAIN_SLOPE_HIGH—Parameter for maximum gain slope. INIT_FRAMES-A parameter that indicates the number of initial frames known to be noise only frames. INIT_NOISE_ENRG_DB-Parameter for initial noise energy value in dB.

INIT_SPCH_ENRG_DB—Parameter for initial voice energy value in dB. MIN_GAIN_HIGH-Parameter for the maximum value of the minimum gain. MIN_GAIN_LOW-A parameter for the minimum value of the minimum gain. NUM_CHAN-A parameter indicating the number of channels. SNR_THLD-A parameter that acts as an SNR threshold. For frames with an SNR greater than this value, the minimum value of the minimum gain, namely MIN_GAIN
_LOW is assigned.

VM_SUM_THLD-A parameter that acts as a voice measurement sum threshold.
Frames with speech measurement sums higher than this are considered to be highly vocalized.

The steps involved in estimating the filtered voice energy are: First
The step is to initialize the filtered speech energy estimate to some suitable value.

[0024]

[Equation 4] The next step is to detect if the current frame is voiced by determining if the total audio measurement exceeds a preselected threshold. If so, the peak channel energy is taken and used as a sample to obtain the filtered speech energy estimate.

[0025]

[Equation 5]

[Equation 6]

[Equation 7] Note that the filter used is a simple leaky integrator (ie first order autoregressive) and different integration constants are used depending on whether the voice energy is increasing or not. The integration is fast if the voice energy is increasing, otherwise it is slow. This makes the filtered estimate smoother and ensures tracking of more reliable peak speech energy in low SNR conditions. If the current frame is not a voiced frame, the filtered voice energy estimate is unchanged from its previous value.

The steps involved in estimating the filtered noise energy are:
The first step is to initialize the filtered noise energy estimate to some suitable value.

[0027]

[Equation 8] The next step is to detect if the current frame is a noise-only frame by update_flag and obtain the total noise energy used as a sample to obtain the filtered noise energy estimate.

[0028]

[Equation 9]

[Equation 10] Note that the filter used is a simple leaky integrator (ie first order autoregressive) and the integration constant is δ. If the current frame is not a noise only frame, the filtered noise energy estimate is unchanged from its previous value.

[0029]   The SNR level for the current frame in dB is given by:

[0030]

[Equation 11] Depending on the SNR level, the min_gain and gain_slope parameters are selected as follows. First, the raw value of the minimum gain is calculated from the SNR level and kept within the limits defined by MIN_GAIN_LOW and MIN_GAIN_HIGH.

[0031]

[Equation 12]

[Equation 13]

[Equation 14] The raw value is then filtered to avoid sudden changes in the min_gain value.

[0032]

[Equation 15]

[Equation 16] Next, the value of gain_slope is calculated as follows.

[0033]

[Equation 17] MIN_GAIN and GAIN_SLOPE are min_gain and gain_s
min_gain and gain_slope depending on the SNR level are used to calculate the gain factor for different channels, using (1), (2) and (3) respectively replaced by the loop.

[Brief description of drawings]

FIG. 1 is a schematic block diagram showing a speech coder for a communication system.

FIG. 2 is a schematic block diagram showing a noise suppression system according to the present invention.

FIG. 3 is a schematic diagram showing the relationship between channel SNR (dB) and channel gain coefficient (dB).

─────────────────────────────────────────────────── ─── Continued front page    (81) Designated countries EP (AT, BE, CH, CY, DE, DK, ES, FI, FR, GB, GR, IE, I T, LU, MC, NL, PT, SE, TR), BR, J P, KR (72) Inventor Ashley, James P.             United States 60565 Ne, Illinois             Iperville Arabian Avenue             1816 F-term (reference) 5K046 AA05 BB01 DD02 DD13 DD25                 5K052 DD02 EE12 EE13 FF05 FF34

Claims (10)

[Claims] 1. A method for suppressing acoustic background noise in a communication system, the method comprising: estimating a signal component and a noise component of an input signal to produce a signal to noise ratio estimate; and at least the signal. Determining a maximum noise attenuation coefficient based on the noise-to-noise ratio estimate, generating a gain function based at least on the maximum noise attenuation coefficient, and generating a noise-suppressed signal for use in the communication system. Applying the gain function to the input signal. 2. The method according to claim 1, wherein the signal component is a voice component. 3. The method according to claim 1, wherein the maximum noise attenuation coefficient is a minimum gain. 4. The method of claim 1, wherein the gain function is a gain for at least one channel. 5. The method of claim 1, wherein the signal component of the input signal is estimated based on filtered peak channel energy. 6. The method of claim 5, wherein the filtered peak channel energy is updated during a period when the input signal exhibits a strong signal content. 7. The method of claim 5, wherein the filtered peak channel energy comprises a maximum channel energy corrected for pre-emphasis gain. 8. The method of claim 7, wherein the maximum channel energy is corrected for pre-emphasis gain using a waveform table. 9. An apparatus for suppressing acoustic background noise in a communication system, the signal-to-noise for estimating a signal component and a noise component of an input signal to generate a signal-to-noise ratio estimation value. A level estimator, an adaptive block for determining a maximum noise attenuation coefficient and a gain slope based at least on the signal to noise ratio estimate, and for generating a noise suppressed signal for use in the communication system, A gain calculator for calculating a gain function applied to the input signal. 10. The apparatus of claim 9, wherein the signal-to-noise level estimator estimates a signal and noise component of the input signal using a channel energy value and a background noise energy value. A device characterized by.