/whisper_api

The objective of this repository is make a basic a functional speech to text api configurable and escalable . The api use a powerful speech to text model, whisper,

Primary LanguagePythonApache License 2.0Apache-2.0

Whisper API

Abstract

Whisper API is a functional ans scalable speech to text API developed using Python and whisper as base. The objective of this repository is give some easy to configure base api to integrate in some special case, for this purpose is necessary to take that we use a client-side pattern (it's possible to change depending of the case). Also, we give the docker container to simplify the test and the deployment, check de package zone.

Table Of Contents

Setup

  1. Download dependencies

    # on Ubuntu or Debian
sudo apt update && sudo apt install ffmpeg  libasound-dev libportaudio2 libportaudiocpp0 portaudio19-dev

# on Arch Linux
sudo pacman -S ffmpeg libasound-dev libportaudio2 libportaudiocpp0 portaudio19-dev 

# on MacOS using Homebrew (https://brew.sh/)
brew install ffmpeg pyaudio

# on Windows using Chocolatey (https://chocolatey.org/)
choco install ffmpeg pyaudio

# on Windows using Scoop (https://scoop.sh/)
scoop install ffmpeg pyaudio

  2. Create a python environment / enable the environment

    conda create --name whisper python=3.10 -y
conda activate whisper
pip install -r requirements.txt

  3. Run

    uvicorn app:app --reload

  • Docker

    docker run -d --name WhisperAPI -p8000:80 ghcr.io/danielsarmiento04/custom_whisper_api:latest

Next Steps

  1. Websocket service, idea

License

This repository is licensed under the Apache 2.0 License.

References

[1] Radford, A., Kim, J. W., Xu, T., Brockman, G., McLeavey, C., & Sutskever, I. (2022). Robust Speech Recognition via Large-Scale Weak Supervision. doi:10.48550/ARXIV.2212.04356

[2] Ramírez, S. FastAPI [Computer software]. https://github.com/tiangolo/fastapi