What's new in Kiosk Satellite 2026.8.19
Multichannel USB microphone arrays often reserve one channel for speech recognition: on a reSpeaker XVF3800, channel 1 carries the call-tuned output with noise suppression and automatic gain control, while channel 2 carries the same voice with lighter processing, the output its maker recommends for wake word and speech engines. Android's mono capture quietly averages every channel together, so the processed channel bleeds into the clean one. This release lets you pick.
New Feature: Microphone channel selection
- A Microphone channel setting under Microphone settings, shown only when the microphone selected under Audio Devices reports more than one channel. Most built-in microphones never show it.
- Picking a channel feeds it alone to wake word detection, the stop word and the audio streamed to Home Assistant for speech to text. Downmix stays the default, so an untouched install captures exactly as before.
- The row appears in the remote admin in the same place and follows the microphone selection live in both UIs.
- A channel the microphone cannot deliver, for example after swapping the array for a simpler microphone, falls back to the downmix rather than going silent.
Recommended setup for the reSpeaker XVF3800
- Set Microphone channel to Channel 2, the output the array reserves for recognition engines.
- Set Capture mode to Raw microphone. The array has already done echo cancellation, noise suppression and gain staging on its own silicon before the audio crosses the USB wire, so Android's phone-call capture path has nothing left to add and can only degrade the stream.
- Re-tune Microphone gain with the Wake Word Tester open, aiming for normal speech around 0.05 from where you actually use the device. Around -10 dB is a sensible starting point; a gain tuned against the old downmix will be too low, since the downmix included the hotter call channel.
- One thing to know: the array's echo canceller only hears what plays through its own speaker output. If Voice Satellite sounds play from the tablet's speaker, the stop word can hear the device's own speech during TTS; routing sounds to a speaker connected to the array puts them back inside its hardware echo cancellation.