Different disturbances interfere with the speaker’s voice signal and without elaborate signal processing, communication would not be possible, neither between humans nor with a machine.
AEC and NR control the two disturbing signal components in a vehicle: loudspeaker-output and driving-noise. Since the source signals of loudspeaker outputs are known, these disturbances can be completely removed from the microphone signal via a two-step process.
Linear echo cancelling is an adaptive algorithm that calculates the impulse responses between loudspeakers and microphones, and estimated echoes are calculated that are then subtracted from the microphone signals. Nonlinear residual echo suppression removes the echoes that remain after linear cancelling, as the human ear is extremely sensitive to echoes of one’s own voice.
For other omnipresent disturbances (driving noise), it can be assumed that this changes slowly over time in a running car and an adaptively calculated noise estimation serves as internal representation of the actual, current noise that is then reduced by the NR algorithm by up to 20 dB.