Echo means that after the sound signal undergoes a series of reflections, it hears the voice of its own speech, which is the echo. Some echoes are necessary, such as music echoes in theaters and rooms with short delays; most echoes are negative, such as hearing yourself talking repeatedly in wired or wireless communications (recall those years when we started When Hei is playing games, if one of them turns on the loudspeaker, his voice will echo back and forth). Therefore, it is very necessary to eliminate the negative influence of echo to the communication system.
For the problem of Acoustic Echo Cancellation (AEC), the most popular algorithm nowadays is the echo cancellation algorithm based on adaptive filtering. This article starts from the two classifications of echo signals and the basic principles of AEC, introduces several classic AEC algorithms and explains their performance.
echo classification
In communication systems, echoes are mainly divided into two categories: circuit echoes and acoustic echoes ( linear echoes and nonlinear echoes )
circuit echo
Circuit echo is usually generated in wired calls. In order to reduce the price of the telephone line between the telephone central office and the telephone user, the connection between the user line adopts a two-wire system; while the connection between the telephone central office adopts a four-wire system (the above two lines It is used to send signals to the client, and the following two lines are used to receive signals from the client). The problem came out. The root cause of the circuit echo is that the two-wire-four-wire impedance of the conversion mixer cannot be completely matched (the reason for the use of different types of wires or load coils is not used), which causes the mixer to receive the voice signal on the line. Lost to the transmission line, an echo signal is generated, allowing the user on the other end to hear his own voice while receiving the signal.
In today's digital communication network, the conversion mixer is integrated with the digital-to-analog converter, but whether it is an analog electronic circuit or a digital electronic circuit, the two-to-four-wire conversion will cause impedance mismatch problems, which will lead to its circuit Echo, affecting the quality of modern communications. Due to the linearity and stability of the circuit echo, circuit echo cancellation can be realized with a simple linear adder . Firstly, the generated echo signal is numerically reversed, and linearly superimposed on the echo signal, and the generated echo signal is canceled to realize the preliminary elimination of the circuit echo. However, due to technical defects, the linear adder cannot completely erase the echo signal, so it is necessary to add a nonlinear processor, which is essentially a switch that blocks the signal. After the residual echo signal is nonlinearly processed, the circuit can be realized. Echo cancellation, or a quiet signal with very little noise. Since the circuit echo signal is linear and stable , it is relatively easy to eliminate it, but this paper mainly studies how to eliminate nonlinear acoustic echo.
acoustic echo
Acoustic echoes can be easily generated in duplex communication systems where microphones and speakers interact. As shown below
Far-end talker-->Far-end microphone-- >Call network----> Near-end speaker--->Near-end microphone -->Call network--> Far-end speaker--->Far-end microphone-- -> Far-end phone --> Near-end phone----> ...... just like this infinite loop,
Detailed explanation: The voice of the far-end speaker is collected by the far-end microphone and transmitted to the communication device. After wireless or wired transmission, it reaches the near-end communication device and is played through the near-end speaker. This sound will be collected by the near-end microphone to form The acoustic echo is transmitted back to the far-end communication device and played through the far-end speaker so that the far-end talker hears his own echo.
Detailed explanation: The voice of the far-end speaker is collected by the far-end microphone and transmitted to the communication device. After wireless or wired transmission, it reaches the near-end communication device and is played through the near-end speaker. This sound will be collected by the near-end microphone to form The acoustic echo is transmitted back to the far-end communication device and played through the far-end speaker so that the far-end talker hears his own echo.
Far-end voice signal : the signal collected by the far-end microphone (speaker voice), which is also equal to the voice played by the near-end speaker, also called reference voice
Near-end speech signal : near-end speaker speech signal
The voice signal received by the near-end microphone : the sound played by the near-end speaker + the voice reflected by the multipath in the room + the voice of the near-end speaker
Far-end mixed echo signal : During the entire conversation, the signal received by the near-end microphone includes the voice signal of the near-end speaker and the voice of the far-end speaker played by the near-end speaker, so that the superimposed voice signal is transmitted to the far-end speaker through the transmission line Playback causes the person at the far end to hear the speech signal he has just detected , the so-called echo.
Acoustic echo signals can be divided into direct echo signals and indirect echo signals according to differences in transmission paths .
- Direct echo (linear echo) : After the near-end speaker plays the voice signal, the echo is directly collected by the near-end microphone; the direct echo is not affected by the environment, and has a great relationship with the distance and position from the speaker to the microphone. The direct echo is therefore a linear signal .
- Indirect echo (non-linear echo) : After the near-end speaker plays the voice signal, the voice signal is collected by the near-end microphone after being reflected by the complex and changeable wall; the size of the indirect echo depends on the room environment, the placement of objects and the attraction of the wall Coefficients and other factors are related, so the indirect echo is a nonlinear signal .
Echo cancellation technology is mainly used in situations such as hands-free telephones and teleconferencing systems .
A complete echo cancellation system includes the following modules:
- Time Delay Estimation (TDE) module
- (Linear) Echo Cancellation (Linear Acoustic Echo Cancellation, AEC) module
- Double-Talk Detect (DTD) module
- Nonlinear Residual Acoustic Echo Suppression (RAES) module
Fundamentals of AEC
Common Methods of Echo Cancellation
- Sound field environment material processing
- Walls and ceilings are replaced with sound-absorbing materials, effectively reducing the reflection of sound,
- Advantages and disadvantages: indirect noise can be suppressed more directly, but direct noise cannot be suppressed; the cost is high
- echo suppressor
- Alternately switching speaker and microphone on and off via a level comparison unit
- Advantages and disadvantages: the output signal of the power amplifier equipment is discontinuous, and the overall effect is not good
- Adaptive echo cancellation (most used)
- Adjust the iterative update coefficient of the filter through an adaptive algorithm, estimate an expected signal, and approach the echo signal passing through the actual echo path, that is, to simulate the echo signal, and then subtract the simulated echo from the mixed signal collected by the microphone to achieve The function of echo cancellation
The most common way to solve AEC problems today is to
Use an adaptive filtering algorithm to adjust the weight vector of the filter, estimate an approximate echo path to approximate the real echo path, and obtain an estimated echo signal , and remove this signal from the mixed signal of pure speech and echo to achieve echo cancellation .
x(n) is the far-end voice, through the unknown echo path w(n) to get the far-end echo voice y(n)=x(n)∗w(n), plus the near-end voice s(n), get Expected signal d(n)=y(n)+s(n). x(n) obtains the estimated echo signal through the adaptive filter
, and subtracts it from the expected signal d(n) to obtain the error signal
. The smaller the value of the error signal, the closer the echo path estimated by the adaptive filtering algorithm is to the actual one. echo path.
The filter uses a specific adaptive algorithm to continuously adjust the weight vector, so that the estimated echo path w^(n) gradually approaches the real echo path w(n). Obviously, in the AEC problem, the choice of adaptive filter plays a very critical role in the performance of echo cancellation.
Reference link: