Audio Architect in acoustic echo cancellation
Acoustic echo occurs in a conferencing system when the far-side speech played in local loudspeakers is picked up by microphones in the near-side room and is transmitted back to the far side. This transmitted signal is a delayed version of the original, which causes the echo.
The received far-side signal does not transfer directly from the speaker to the microphone, but is subject to the artifacts of the room. This may include differing signal paths causing reverb, frequency filtering and attenuation. These effects are the result of the transfer function of the room. This transfer function is also dynamic as objects in the room move or the microphone moves position.
To subtract the unwanted signal correctly, the Acoustic Echo Cancellation (AEC) processor needs to simulate the dynamic room transfer function. It can then apply that transfer function to the received signal and correctly subtract the modified original signal.
Each Soundweb London AEC input card consists of four AEC input channels.
Each channel offers the following features:
● Independent 20Hz - 8kHz algorithm
● Individual AEC references
● Automatic Gain Control
● Noise Cancellation
● Adaptive (Speech Passing) Non-Linear Processing
● Extremely fast convergence rates of 49dB/s
EXAMPLE: BASIC CONFERENCING WITH LOCAL SOUND REINFORCEMENT
This example shows four microphones feeding the audio to the far side via a Telephone Hybrid card as well as feeding the local speakers for local sound reinforcement. Signal mixing is performed using the Gain Sharing Automixer processing object. The best method for this type of design incorporates a mix-minus setup to maintain proper gain structure, and to prevent the speaker directly above the person talking from transmitting a room-colored copy that will re-enter the open microphone and be transmitted to the far side along with the original voice signal.
The design displayed here shows both the far-side and near-side signals feeding the local room speakers. This design works, but as explained previously, the AEC algorithm will not perform to its full potential.
If the Reference is moved to the same location as in the previous ‘No Local Sound Reinforcement’ example, it will satisfy the rule of placing the Reference as close as possible to the speaker output, but in doing so the Reference will be fed with a mix of both the near-side and far-side signals.
Since the Reference signal is the signal to be removed from the input audio path, the AEC algorithm will cancel the microphone signal at the AEC Input Card. Since the input microphone signal path is being fed to the far-side as well as the local loudspeakers, listeners at the far side would not be able to hear the microphone signal. Only a portion of the microphone signal is cancelled because of the Voice Activity Detection state. The Voice Activity Detection processing determines whether the audio is speech or silence / background noise. This would cause the microphone signal to distort both locally and at the far side.
To solve this, another set of High / Low Pass processing objects and a Parametric EQ are utilized, in order to provide the Reference the same signal as the room speaker. This results in only the far-side signal being Referenced and removed while still feeding a mix of both near-side and far-side audio to the room loudspeakers.
It is essential that the same settings are maintained in both signal paths. In particular, care must be taken that any non-linear processing such as compression or limiting employed on the speaker output signal, is also applied to the Reference signal. The Copy Parameter Values feature should be employed to ensure that the settings are identical.