stt-server

Minimal Whisper-based speech-to-text HTTP server.

Usage

Pre-built docker container: rctl/stt-server:latest

docker run rctl/stt-server:latest -p 8081:8000

What It Does

• Loads an OpenAI Whisper model on startup (turbo by default).
• Exposes a single transcription endpoint: POST /transcribe.
• Accepts raw PCM s16le mono audio in the request body.
• Returns Whisper JSON output (including text and segments timing).

API

• GET / returns ok
• POST /transcribe
  • Headers:
    • X-Sample-Rate: sample rate (default 16000)
    • X-Lang: language code (default en)
  • Body:
    • raw int16 PCM bytes
  • Response:
    • Whisper transcription JSON

Run

pip install -r requirements.txt
python main.py --host 0.0.0.0 --port 8000 --model turbo

Optional flags:

• --device cuda|cpu
• --no-fp16
• --debug (writes incoming audio samples as *.wav)

Quick Test

curl -X POST "http://localhost:8000/transcribe" \
  -H "X-Sample-Rate: 16000" \
  -H "X-Lang: en" \
  --data-binary @sample.pcm