Enemies of Voice

One pet peeve of my very much needed protective case for my S8 is that it creates a channel around the microphone inlets. Channels can have a profound effects on voice interaction by changing the pitch, introducing reverberation, or even obscuring the microphones from different angels.

The three basic enemies of voice interaction are:

Low Signal to Noise Ratio (SNR)

Reverberation

Acoustic Echo

Low SNR can be caused by distance — sound signal follows a power law. For every meter distance between voice and microphone, you halve the power of the signal (e.g. 70 dB at 1 m will be 60 dB at 2 m). The other cause can be noise. Lots of noise, especially speech, can make things difficult.

Reverb is when the main signal bounces of a surface and the echo of the signal hits just after the main signal, creating a slur in the signal. This makes it difficult for speech recognizers to decipher the meaning.

Acoustic Echo is when the output from the device feeds back into the microphones, essentially creating a lot of noise that makes it difficult for the device to hear anything else. Countering this is AEC — Acoustic Echo Cancellation — filters out the output signal. Another strategy is then to duck audio as soon as a wake word is heard.

Unintended channels can also wreak havoc on far field algorithms, such as the issue with a cellphone case. Surface bounce and close proximity to reflective objects also cause problems. Beware the enemies of voice!