Moving Frontends

Fig. 21 Moving system considerations

Moving systems, such as service robots, or devices that are moved or covered, like kitchen appliances, have a special requirement: they need fast acoustic adaptation. Modern advanced audio systems must therefore be tuned to adapt quickly enough, as it becomes a race between movement speed and adaptation time. During rapid changes, beamforming may initially listen in the wrong direction and Echo cancellation might not reach the target SNR.

Adaptation Times of Advanced Audio Frontends

• Echo Canceller (AEC): (~300 ms - 1 second adaptation time)

• Denoise: Noise / Frequency response weights (2-5 sec)

• Beamforming/Directional filtering (few 100 ms).

End-to-End AI/DNN-Based Beamformer with AEC

• 2-5 Frames (32-150 ms) adaptation time (full system).

Hint: Since AEC can only adapt while audio is being played, it may be useful to play short jingles to enable adaptation, or to expect AEC residuals during the first second of playback and harden the system accordingly. If possible, avoid placing the microphones and loudspeakers on the fastest moving parts (e.g. arms, manipulators).