I've tried a couple of the releases (latest 0.0.3) on a MacBook Pro M4, both using the built-in recording using the mic and by dragging and dropping a (short, normalized, mono) voice recording in English.

No matter what I do, I always get the "No speech detected in audio". I've also tried to adjust the "no speech threshold" and set it all the way to 0.0, but that doesn't have any impact.

I can replay the recording (and hear my voice clearly), so it's not the recording itself that is wrong.

Are there limits to the length of the recording (does it need to be at least some length)?

Is there a known-good recording I can test with, in case it's something to do with the way I record the audio.

Is there a way to get some debug info on what the model is doing?