How can you get your Voice Assist pipeline to generate fewer errors? This is dumb, but it works.... just name things with more common, distinct words. Whisper seems to have more problems recognizing less common words, which makes sense from a ML training perspective, I guess. Those words would probably pop up more in the training data.
For the speakerphone I'm using, see: • Voice Control and Wake Word Detection on U...
The espresso machine is plugged into a Sonoff S31 that I converted to run ESPHome.
• Reflash Sonoff S31 to ESPHome with teardow...
• Making a Pogo Programmer