Insanely Fast Whisper API Deployable to Fly GPU

github.com

24 pointsindigodaddy4 comments

Beautiful, would love a way to deploy this easily on Huggingface Inference Endpoints.

The need for a custom handler.py has always stopped me from doing so.

Is there a commercially available whisper api or you have to use it locally only ?

OpenAI's speech to text API is the official Whisper implementation. It's expensive.

Many small providers will host Whisper as an endpoint (Replicate for example).

Groq has Whisper in a private beta; if they open it up for commercial usage, cheap and fast, that'll be a game changer.

NLP Cloud is a good example of a smaller provider that proposes a good Whisper implementation (no downtimes and relatively cheap): https://nlpcloud.com/nlp-automatic-speech-recognition-speech...