tiny-openai-whisper-api

Based on https://github.com/morioka/tiny-openai-whisper-api, using whisper-mlx instead of openai-whisper

OpenAI Whisper API-style local server, runnig on FastAPI. This is for companies behind proxies or security firewalls.

This API will be compatible with OpenAI Whisper (speech to text) API. See also Create transcription - API Reference - OpenAI API.

Some of code has been copied from whisper-ui

Setup

This was built & tested on Python 3.10.8, Ubutu20.04/WSL2 but should also work on Python 3.9+.

sudo apt install ffmpeg
pip install fastapi python-multipart pydantic uvicorn ffmpeg-python
# or pip install -r requirements.txt

or

docker compose build

Usage

server

export PYTHONPATH=.
uvicorn main:app --host 0.0.0.0

or

docker compose up

client

note: Authorization header is ignored.

example 1: typical usecase, identical to OpenAI Whisper API example

curl http://127.0.0.1:8000/v1/audio/transcriptions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: multipart/form-data" \
  -F model="whisper-1" \
  -F file="@/path/to/file/openai.mp3"

example 2: set the output format as text, described in quickstart.

curl http://127.0.0.1:8000/v1/audio/transcriptions \
  -H "Content-Type: multipart/form-data" \
  -F model="whisper-1" \
  -F file="@/path/to/file/openai.mp3" \
  -F response_format=text

License

Whisper is licensed under MIT. Everything else by morioka is licensed under MIT.