CN101286763A - An Effective Echo Suppressor - Google Patents

Abstract

An effective echo suppressor of the present invention includes two adaptive gain units G _r (RSR) and G _n (NSR), an adaptive zero filter unit A ₁ (z) and an adaptive pole filter unit A ₂ (z); where, the gain unit G _r (RSR) is controlled by RSR (ratio of residual echo level to signal level); the gain unit G _n (NSR) is controlled by NSR (noise signal level and current signal (Tx ) level ratio) control; filter unit A ₁ (z) is converted by LSF ₁ , LSF ₁ is based on the first modification of LSF _Tx (linear spectral frequency of Tx signal); filter unit A ₂ (z) is converted by LSF ₂ conversion, LSF ₂ is based on another modification of LSF _Tx . This kind of suppressor can not only suppress the echo but also suppress the background noise, and does not produce any "discontinuity" in practical application, so that no obvious speech distortion will be heard. It is an effective and easy-to-implement method. echo suppressor.

Description

An Effective Echo Suppressor

technical field

The invention relates to a telecommunication device, more precisely it introduces a method for efficient echo cancellation and suppression.

Background technique

echo canceller

An echo canceller can be a device or a piece of software that cancels echo signals by using a reference signal. The reference signal is also called a received signal (Rx signal). The echo signal is also called the echo return signal, which is mixed in the transmission signal (Tx signal). There are two main types of echo cancellers: one is called an acoustic echo canceller (AEC) and the other is called a linear echo canceller (LEC). Obviously, the acoustic echo canceller is used to cancel the acoustic echo, and the linear echo canceller is used to cancel the linear echo.

The reason for the linear echo is due to the imbalance between the hybrid circuit and the impedance when the two-to-four-wire signal is transformed. Acoustic echo occurs in telecommunication equipment, it is produced due to the combined echo of voice being transmitted back to the receiving party while providing a full-duplex connection; this problem also occurs in teleconferencing - how to set up two or more people Eliminate noise from other people while talking on the bridge. During voice communication, some echo is acceptable, but users usually don't want to hear their own speech; that is, echo delayed by the round-trip time of the system.

The echo canceller is usually an adaptive filter combined with a least mean square (LMS) algorithm that produces an echo replica that approximates the echo signal. There is also a subtraction between the echo return signal and the echo homemade signal for eliminating the echo return signal. For various reasons, the replica signal cannot completely replicate the echo return signal, so there will be some residual echo in the transmitted signal. An echo signal suppressor is a device that can effectively reduce or eliminate echoes, and it is especially used to reduce or eliminate residual echoes in signals processed by an echo canceller. Due to the inherent problems with the echo canceller, various solutions have relied heavily on adding an echo signal suppressor.

echo signal suppressor

An echo suppressor can be a device or a piece of software that effectively reduces (residual) echo energy without significantly distorting non-echo speech signals. Although an echo suppressor can work independently of an echo canceller, it is usually used in conjunction with an echo canceller. Echo suppressors can not only effectively suppress residual echo energy, but also eliminate background noise energy with the help of some useful parameters. Echo suppression can be considered an independent function or just a part of the echo cancellation system.

Most of the existing echo suppression methods are realized through the following ways:

· Spectrum echo suppression is one of the commonly used methods. This method is relatively simple, it converts the time domain signal into frequency domain through FFT operation. In practice, it requires careful tuning of all parameters to avoid possible "musical noise" before the corrected spectrum is converted back to the time domain.

· Another more common method is called the nonlinear processor method (NLP), which replaces the residual echo signal with random noise, or cuts the reverse transmission of the signal when only one person is speaking. The NLP method is simple, but requires more accurate detection of double-ended sound (Double Talk) and residual echo. This method usually produces "discontinuities" in noisy environments due to the difficulty of obtaining a more accurate double-ended sound detector or a more perfect residual echo detection method.

Contents of the invention

The purpose of the present invention is to overcome the shortcomings of the above-mentioned existing echo signal suppression methods, and propose a method that can suppress both residual echo and background noise, and can even be applicable to double-ended sound environments, and in practical application An effective echo suppressor that does not introduce any "discontinuities" and does not audibly distort the speech. It is highly effective and easy to implement.

An effective echo suppressor of the present invention includes two adaptive gain factors G _r (RSR) and G _n (NSR), an adaptive zero filter unit A ₁ (z) and an adaptive pole filter unit A ₂ (z), characterized by:

The gain factor G _r (RSR) is controlled by RSR;

The gain factor G _n (NSR) is controlled by NSR;

The filter unit A ₁ (z) is converted from LSF ₁ ; LSF ₁ is the first modification based on the initial setting LSF _Tx ;

The filtering unit A ₂ (z) is converted from LSF ₂ ; LSF ₂ is the second modification based on the initial setting LSF _Tx .

The RSR is the ratio of the residual echo level to the signal level; the signal level refers to the Tx signal level of the current frame or sub-frame; the residual echo level is the product of RRR and the current received signal (Rx) level Calculated.

The RRR is the average ratio of the residual echo level to the received signal (Rx) level.

The NSR is the ratio of the background noise level to the current signal (Tx) level.

The LSF refers to Line Spectral Frequencies.

The above initial setting of LSF _Tx before modification is obtained based on the LPC analysis of the Tx signal.

The first modification of the initial setting LSF _Tx and the second modification of the initial setting LSF _Tx are controlled by RSR, NSR, LSF _echo and LSF _noise .

The estimation of the LSF _echo can be obtained by means of the relationship between LSF _Rx and LSF _Tx in the pure residual echo region of the Tx signal; here, the LSF _Rx is obtained based on the LPC analysis of the Rx signal.

The LSF _noise of the background noise is obtained based on the LPC analysis in the background noise area of the Tx signal.

The present invention has the advantages of not only suppressing residual echo but also suppressing background noise, even applicable to double-ended sound environment, and does not introduce any "discontinuity" in practical application, so that no obvious speech distortion can be heard . It is an effective echo suppressor with high utility and easy implementation.

Description of drawings

FIG. 1 is a block diagram of an example of an acoustic echo canceller (AEC) 112;

FIG. 2 is a block diagram of an example of a linear echo canceller (LEC) 209;

Fig. 3 is a schematic diagram of the spectrum envelope of a double-ended speech signal, wherein the residual echo signal is mixed with the non-echo speech signal;

4 is a schematic diagram of a reference signal Rx waveform;

Fig. 5 is a schematic diagram of the waveform of the input (residual) signal Tx before the echo suppressor;

FIG. 6 is a schematic waveform diagram of the output signal Tx after the echo suppressor.

Detailed ways

The present invention is further described as follows in conjunction with accompanying drawing:

Information specific to this type of echo suppressor is described below. But those skilled in the art will understand that the invention can be applied to some different occasions in combination with different algorithms. Some details known to those skilled in the industry will not be discussed here, so as not to obscure the key points of the present invention.

The contents introduced in the accompanying drawings and related descriptions herein are only some specific application examples of the present invention. For the sake of brevity, other examples of the same application principles involved in the present invention will not be illustrated and described in detail here.

FIG. 1 is an example of an acoustic echo canceller (AEC) system 112 . The echo signal 111 is transmitted from the speaker 110 and returns to the microphone 101 . The signal sent to the loudspeaker is called the received signal 109 or the reference signal of the acoustic echo canceller 103 and the echo suppressor 105 . The signal entering the Acoustic Echo Canceller (AEC) and coming out of the Microphone (MIC) 101 is called the Transmit Signal 102 (Tx Signal), this transmit signal has an echo return signal (originating from the speaker). An acoustic echo canceller (AEC) 103 and an echo suppressor 105 are responsible for canceling or suppressing echo signals. The acoustic echo canceller (AEC) cancels or suppresses the echo return signal by means of the reference signal Rx 109 to generate a replica signal similar to the echo return signal. The residual echo signal 104 is further suppressed by a post-processing module 105, which may be an echo suppressor or a nonlinear processor (NLP).

FIG. 2 is an example of a linear echo canceller (LEC) system 209 . This is a typical example where the echo signal 211 is reflected from the telephone mixer 210; the transmit signal Tx1 201 contains the echo return signal; An echo return signal 211 is canceled or suppressed. LEC works similarly to AEC. One of the main differences is the difference in the echo path. Also the range of echo hysteresis may vary. The echo canceller uses the reference signal Rx 208 to generate a replica signal similar to the echo return signal, thereby canceling or suppressing the echo return signal. The residual echo signal Tx2 203 will be further suppressed by a post-processing module 204, which can be an echo suppressor or a nonlinear processor (NLP).

FIG. 3 is a schematic diagram of the signal spectral envelope for an example of double-ended speech in which the residual echo signal is mixed with the non-echoed speech signal. Shown at 302 is the spectral envelope of the mixed signal. 301 is the spectral envelope assuming an echo-free speech signal. In most cases, the residual signal formant 303 is smaller than the speech formant 304 . The residual echo in the double-speech region is more difficult to suppress if it is mixed with the speech signal than the residual signal existing alone; this is because the residual echo signal must be suppressed without distorting the speech signal .

Figures 4-6 present three signal examples: Figure 4 represents the Rx signal (reference signal) 401, Figure 5 represents the Tx signal 402 before the input (residual) echo suppressor (202 in Figure 2), and Figure 6 is the Tx signal 406 output from the (residual) echo suppressor (204 in Figure 2). As shown in FIG. 5 , 403 is a double-ended speech area. An effective (residual) echo suppressor can remove both residual echo signal 404 and background noise 405 under normal operation.

The present invention proposes a highly effective and easy-to-implement echo suppression method. It is implemented by a gain-controlled filtering process, which can be explained by the following filtering model:

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; NSR</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; RSR</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&CenterDot;</span>\frac{{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

Among them, G _n () is a gain, and this gain is a function of NSR (or SNR), and NSR is defined as the ratio of the background noise level to the signal level. The value of NSR can be obtained by measuring the Tx signal before the echo suppressor combined with VAD (Voice Activity Detection) information. G _r () is also a gain. This gain is a function of FSR. RSR is defined as the estimate of the ratio of the residual echo level to the signal level in the Tx signal. This estimate is more complicated and will be explained in detail later in this article. Both A ₁ (z) and A ₂ (z) are linear predictors, which are composed of LPC coefficients, and the LPC coefficients are converted from LSF ₁ and LSF ₂ , where LSF refers to the line spectral frequency (Line Spectral Frequencies ). Both LPC coefficients and LSF are well-known parameters in the field of speech signal processing, and they are often used to represent spectral envelopes. LSF ₁ is derived from the first modification of LSF _Tx , wherein LSF _Tx is calculated based on LPC analysis of the Tx signal; LSF ₂ is derived from the second modification of LSF _Tx . The modification of the LSF is controlled by the parameters SNR, RSR and another set _of LSF _Rx calculated from the LPC analysis of the Rx signal.

In equation (1), the gain usually does not drop to 0, but is small enough in the pure echo region that the pure echo cannot be heard; the main contribution of the gain is that it can significantly reduce Reduce echo or noise energy. In normal speech or double-speech regions, the gain factor is usually smaller than 1 and depends on the parameters NSR and RSR. Since the change of NSR and RSR is relatively smooth and slow, and the change of gain factor is consistent with it, "discontinuity" can be avoided.

The LPC filter units A ₁ (z) and A ₂ (z) in equation (1) are mainly used to suppress the residual echo formant in the double-ended speech area (as shown in Figure 3) or reduce the noise or spectrum in the low SNR speech area magnitude. Since the parameter changes of the LPC filtering units A ₁ (z) and A ₂ (z) are relatively smooth and slow, "discontinuity" is avoided, so that no obvious speech distortion can be heard.

The foregoing content illustrates the basic principles of the present invention. Hereinafter, the present invention will be explained in detail.

• An estimate of NSR (or SNR), which is defined as the ratio of the background noise level to the current Tx signal level. This parameter can be determined by means of a usual method. Background noise refers to the recent average background noise level when only background noise is present in the Tx signal. The signal level refers to the current frame or sub-frame signal level of the Tx signal. When only background noise is present, the NSR value is approximately 1, and approximately 0 dB in the dB domain. In the voice area, the NSR value should be less than 1.

·Residual echo signal detection refers to detecting most of the residual echo signal area under the condition that only residual echo signal and noise exist. This detection unit need not be precise as it will only be used to estimate the average loss of echo energy compared to the Rx signal energy. After the hysteresis between the Rx signal and the echo return signal is detected, both the Rx signal and the residual return signal have been synchronized within the echo canceller. If double-talk is not present, the energy reduction of the residual echo signal after the basic echo canceller is very significant compared to the energy of the original Rx signal. This information can be used to detect most residual echo signals.

• An estimate of the RSR, which is defined as the ratio of the residual echo level to the current Tx signal level. The signal level still refers to the current frame or sub-frame signal level of the Tx signal. Computing the residual echo level is more complicated. If there is no residual echo, the residual echo level is 0. First, the average energy loss (energy reduction) of the residual echo is estimated in the corresponding region only in the presence of the residual echo, which is defined as the residual echo level compared with the corresponding received signal level (Rx signal A ratio of level). Energy levels can be expressed directly or in the dB domain. The calculation expression of its average ratio (or moving average) is as follows:

Therefore, the current residual echo energy level can be estimated by the following formula:

Current residual echo energy level = (RRR) (current Rx signal energy level) (3)

According to the above formula, the current residual echo energy level can be estimated even in the double-ended speech area. Therefore, RSR can be calculated according to the following formula:

According to the above formula, the RSR value is about 1 in the pure residual echo area and less than 1 in the double-ended speech area.

Gain _Gn (NSR), which can be a linear or non-linear function of the parameter NSR. The following is an example of a linear function:

G _n (NSR) = 1-C _n NSR (5)

Among them, C _n is a constant: 0<C _n <1

• Gain G _r (RSR), which can be a linear or non-linear function of the parameter RRR. The following is an example of a linear function:

G _r (RSR)＝1-C _r RSR (6)

Among them, G _r is a constant: 0<C _r <1

• The LSF of the Tx signal, expressed as LSF _Tx (i), i=0, 1, . . . , M-1; its estimation is based on the LPC analysis of the Tx signal. Typical values for the order (M) are around 10 for narrowband signals at a sampling rate of 8kHz.

The LSF of the noise signal, expressed as LSF _noise (i), i=0, 1, ..., M-1; its estimation is based on the average (or moving average) of the LSF _Tx (i) in the background noise region of the Tx signal ).

• LSF of the Rx signal, expressed as LSF _Rx (i), i = 0, 1, . . . , M-1; its estimation is based on the LPC analysis of the Rx signal.

(Residual) echo signal LSF, expressed as LSF _echo (i), i=0, 1, ..., M-1; when the residual echo signal is mixed in the voice signal, LSF _echo (i) Estimates within the region are more difficult. For example, LSF _echo (i) can be derived by extrapolating with LSF _Rx (i). First, the imputation factor P(i) is estimated; it is the recent average ratio (or moving average ratio) between LSF _Rx (i) and LSF _Tx (i) in the pure residual echo region:

Then, the current LSF _echo (i) value of the residual echo is estimated according to the following formula:

LSF _echo (i) = P(i) · LSF _Rx (i), i = 0, 1, ..., M-1 (8)

where LSF _Rx (i) is the current linear spectral frequency of the Rx signal.

·LPC predictors A ₁ (z) and A ₂ (z) are converted from LSF ₁ (i) and LSF ₂ (i) respectively, where i=0, 1, . . . , M-1. Both LSF ₁ (i) and LSF ₂ (i) estimates are based on modifications of LSF _Tx (i). The modification is mainly affected by LSF _echo (i), LSF _noise (i), NSR, and RSR. For example, LSF ₁ (i) and LSF ₂ (i) can be formed as follows:

LSF ₁ (i) = λ ₁ LSF _Tx (i) + β LSF _echo (i) + α LSF _noise (i),

i=0,1,....,M-1 (9)

LSF ₂ (i)=λ ₂ ·[LSF _Tx (i)-β·LSF _echo (i)-α·LSF _noise (i)],

i=0,1,....,M-1 (10)

in:

β = C _β RSR, (11)

α=C _α NSR, (12)

Both C _α and C _β are constants; their values are both greater than 0 but much less than 1. λ ₁ and λ ₂ are determined in the following way:

λ ₁ =1-β-α, (13)

${\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 14</span>}<span\; class="notranslate">\; )</span>$

· EC filtering divergence protection refers to avoiding the Tx signal level after the echo canceller and the echo suppressor to be greater than the level before the echo canceller (remarked as Tx1 signal). It can be ensured that the output energy of the echo suppressor is less than or equal to the energy of the Tx1 signal by simply adjusting the two gain factors C _n and C _r in equation (1).

The present invention can be extended to some other specific forms and applied on the premise of not departing from the essence. The specific examples herein are merely exemplary of the present invention, and the present invention is not limited to these examples. The above contents are only some illustrations of the present invention, and the following claims are the essential contents of the present invention. The following claims include various interpretations and changes within the scope of equivalents thereto.

Claims (9)

1. an effective echo suppressor includes two adaptive gain factor G _r(RSR) and G _n(NSR), self adaptation filter unit at zero point A ₁(z) and an adaptive pole filter unit A ₂(z), it is characterized in that:

Gain factor G _r(RSR) control by RSR;

Gain factor G _n(NSR) control by NSR;

Filter unit A ₁(z) by LSF ₁Convert; LSF ₁Be based on initial setting up LSF _TxFirst revision;

Filter unit A ₂(z) by LSF ₂Convert; LSF ₂Be based on initial setting up LSF _TxSecond revision.

2. a kind of effective echo suppressor as claimed in claim 1 is characterized in that described RSR is the ratio of residual echo level and signal level; Signal level is meant the Tx signal level of present frame or subframe; The residual echo level is calculated by the product of RRR and current received signal (Rx) level.

3. a kind of effective echo suppressor as claimed in claim 2 is characterized in that described RRR is the mean ratio of residual echo level and received signal (Rx) level.

4. a kind of effective echo suppressor as claimed in claim 1 is characterized in that described NSR is the ratio of background-noise level and current demand signal (Tx) level.

5. a kind of effective echo suppressor as claimed in claim 1 is characterized in that described LSF is meant linear spectral (LineSpectral Frequencies) frequently.

6. a kind of effective echo suppressor as claimed in claim 1, the initial setting up LSF before it is characterized in that revising _TxThe lpc analysis that is based on the Tx signal draws.

7. a kind of effective echo suppressor as claimed in claim 1 is characterized in that described initial setting up LSF _TxFirst revision and initial setting up LSF _TxSecond revision be subjected to RSR, NSR, LSF _EchoAnd LSF _NoisControl.

8. a kind of effective echo suppressor as claimed in claim 7 is characterized in that described LSF _EchoEstimation can be by LSF in the pure residual echo district of Tx signal _RxAnd LSF _TxBetween relation draw; Here LSF _RxThe lpc analysis that is based on the Rx signal draws.

9. the described a kind of effective echo suppressor of claim 7 is characterized in that the LSF of described background noise _NoisThe lpc analysis that is based on Tx signal background noise range draws.

Images