Better ASR #3
Description
There is a asr model that has been made for audio <30 seconds called moonshine. We should detected if the input audio is <30 seconds, and if it is, use moonshine, otherwise use whisper.
For our whisper integration, we should use whisper cpp wasm bindings similar to #1 for a speed increase, since it is pretty slow right now. Ideally we would also be able to use a larger model than whisper base (whisper distil?), since whisper base's performance has been lack luster so far in testing