A server to abstract the available whisper implementations and serve through a common websocket interface. Use it with speech2text-app.

First install ffmpeg in the appropriate way for your platform. Check https://github.com/kkroening/ffmpeg-python for more info.

Create a new venv, activate it and install the base dependencies:

pip install -r requirements.txt

Create a copy of the .env.template file as .env and edit the settings as desired.

Now you choose an engine and set it in the env file:

• whisper: The original Whisper from OpenAI
• whisper.cpp: Port of OpenAI's Whisper model in C/C++ (https://github.com/ggerganov/whisper.cpp)
• openai: Uses the OpenAI online API

And finish the configuration for each:

Install the extra packages.

• Only if you have a NVIDIA GPU: pip install -r requirements-whisper-nvidia.txt
• Or else: pip install -r requirements-whisper.txt

Unix only

You'll probably need to install compilation related packages before.

Install the extra packages:

pip install -r requirements-whispercpp.txt

Uses the online OpenAI API platform. You'll need to set up an account, get an API key and set it in the .env file.

There's a test Gradio interface available. Set USE_GRADIO=1 in the .env file and install extra packages:

pip install -r requirements-gradio.txt

Run this to start the server:

uvicorn main:app --env-file .env --port 9748 --host 0.0.0.0

The server will be available at http://127.0.0.1:9748/. If you enabled Gradio, the test interface will be available at http://127.0.0.1:9748/gradio/.