That’s seriously impressive — getting wav2vec2-bert running with IREE in Rust must’ve taken a lot of tinkering. I’m really interested in how efficient the inference is, especially since Rust is so performance-focused. Are you planning to use this for any kind of real-time transcription?

Also, if you’re testing on mobile or need a temporary number for voice input/output scenarios, TextNow might be worth checking out. Some developers use it during testing phases — reviews here https://textnow.pissedconsumer.com/review.html are mixed, but it can be useful for lightweight communication or mock user scenarios.