Each channel of Acoustic Echo Cancellation (AEC) has 2 inputs and one output.
- AEC Input is the audio signal from the microphone
- AEC Reference is the audio signal that is being sent to the loudspeakers.
- AEC Output is the microphone audio after the loudspeaker audio has been removed.
The AEC object is made up of multiple, internal components.
- The AEC Adaptive Filter is a programmable filter used to remove the loudspeaker audio from the microphone signal.
- The AEC Algorithm compares the AEC Reference to the output of the Adaptive Filter in order to generate and refine a Room Impulse Response (RIR) that matches the teleconference room. That Room Impulse Response is used to program the AEC Adaptive Filter.
- Non-Linear Processing (NLP) eliminates any residual loudspeaker audio present in the microphone signal after the microphone audio has been processed by the AEC Adaptive Filter. NLP also provides very fast, but lower quality echo cancellation during times when the AEC Algorithm has not fully converged, and the AEC Adaptive Filter is not yet able to eliminate the loudspeaker audio.
- Noise Reduction (NR) eliminates steady-state noise from the microphone signal. Examples of steady-state noise in a conferencing environment include fans in projectors and laptops, as well as air handler noise from the HVAC system.
- Comfort Noise (CN) provides filtered, low-level noise, which can be added to the microphone signal to prevent the far end from hearing complete silence when the near end talkers are not speaking. Without Comfort Noise, far end participants are often unsure whether the near end is still connected to the teleconference.
- DoubleTalk Detection (DTD) prevents the AEC Algorithm from trying to refine the Room Impulse response during times when it is impossible to develop a valid Room Impulse Response. When there is no audio coming from the far end, it is impossible to generate a valid Room Impulse Response, so the AEC algorithm is paused. Similarly, the AEC Algorithm is also paused when the near end is talking.