Unlock Beamforming Capabilities

[that1guy]

Summary

The Satellite1 features four onboard microphones and a powerful XMOS processor. We should leverage these capabilities to implement beamforming, enabling the device to perform room separation. This enhancement will allow Satellite1 to distinguish human voice commands from background noise, such as a TV or loud conversations.

Motivation

To compete effectively with major voice assistant brands, the Satellite1 must perform reliably in noisy environments.

[TheBToby]

Even though I'm not a tech expert, the motivation described as part of this enhancement request sounds great. I'm actually struggling a lot with the Satellite1 hardware in noisy environments (e.g. music, children playing, cooking, noise from the outside, etc.). Compared to the Sonos Move standing right beside the Satellite1, wake word detection and voice recognition are far behind.
When listening to the audio file before passing it through Whisper, the voice sounds very distant and "muffled" when not in a quiet environment. Using the same Voice Assistant pipeline with the microphone from my iPhone or laptop, or when standing very close to the Satellite1 works great, though.

[that1guy]

Hi @TheBToby, thanks for your feedback! We completely agree, the Sat1 (and other XMOS projects) is the most affected by arbitrary background noise.

That said, the “muffled” or “distant” sound you’re hearing from the microphones isn’t exactly the core issue. Audio that sounds natural to human ears doesn’t always perform well with a wake word model or STT engines. That’s why the recordings don’t have the same studio quality clarity you might expect.

We definitely need more advanced audio processing algorithms to handle background noise more effectively. The good news is that the processing logic is open source here, and we’re actively exploring improvements.