Use OpenAI API for ASR

[lee-b]

It would be really great to support the OpenAI API for ASR as well as TTS. It would let people choose their own way of running ASR, and where to run it.

In particular, there are OpenAI compatible inference servers for Parakeet, which is significantly better (at least for non-Americans, and with noisey environments) than whisper (even whisper v3 turbo etc.).

By supporting OpenAI API's for this, it would immediately add support for Parakeet and perhaps other engines, and also other acceleration frameworks (MLX, Vulkan, whatever) without specific work within Vocalis.

[Lex-au]

Hi @lee-b

To confirm, you mean an OpenAI-compatible endpoint specifically for the transcription aspect? The state management is fairly complex as is; however, this is something I will look into.

In the interim, you can adjust the vad threshold manually (in the .env) to try and account for false triggers from background noise.