feat: Auto-detect and convert audio file formats

[augustebaum]

I tried to transcribe a .m4a file and got this error:

2025-08-27T12:34:19.444932Z ERROR Failed to transcribe <my-file>: Ill-formed WAVE file: no RIFF tag found

It took me a second to realize that vosk only seems to accept .wav files. It would be super convenient to automatically do the conversion for me, or at least detect that the file format is wrong and have a clearer error message.

[augustebaum]

Note that vosk-transcriber, the Python wrapper around Vosk, seems to accept more file formats; I haven't yet checked if they added extra logic to convert.

[guttermonk]

This would be a nice feature. When I tried using WhisperNow, I couldn't get great performance on NixOS, since it was primarily written in Python and it would frequently have to download, unzip, and build when you ran it.

It there's a way to do this without sacrificing performance, that would be cool. On a side note, I think there's some opportunity to speed up the Jambi's load time by deferring the model check if it already has one. My goal is to get it to 3 seconds or less, but I need to look more into this.