whisper.cpp / README.md
Arseniy Savin
Implement initial setup for Whisper.cpp FastAPI application with Docker support. Added Dockerfile, main application logic, requirements, and updated README with usage instructions and environment variables.
5aea4af
|
Raw History Blame Contribute Delete
779 Bytes
metadata
title: Whisper.cpp
emoji: 🦀
colorFrom: indigo
colorTo: blue
sdk: docker
app_port: 7860

Minimal REST API for Whisper.cpp on free CPU Spaces.

Endpoints

  • GET /health: liveness check
  • POST /v1/transcribe: multipart file upload (audio/video). Returns JSON with text and per-segment metadata.

Quick use

curl -s -X POST \
  -F "file=@sample.mp3" \
  https://<your-space-subdomain>.hf.space/v1/transcribe | jq

Environment

  • WHISPER_MODEL: model name (e.g. tiny.en, base.en, small). Default tiny.en.
  • WHISPER_THREADS: number of CPU threads to use. Default 4.

Notes

  • Uses pywhispercpp wheels (no compile) + ffmpeg to handle non-WAV inputs.
  • Weights cache in ephemeral storage. For persistence, add a storage upgrade and set HF_HOME=/data/.huggingface.