MedASR Transcription Server

OpenAI-compatible speech-to-text transcription server using google/medasr model.

Features

• OpenAI API compatible endpoint (POST /v1/audio/transcriptions)
• Supports common audio formats: mp3, wav, flac, m4a, ogg, webm, mp4
• Model loaded once at startup and cached in memory (no reloading per request)
• GPU acceleration support (auto-detected)
• YAML configuration
• Audio processed in-memory (no temporary files on disk)

Installation

# Create virtual environment (recommended)
python -m venv venv
./venv/scripts/activate

# Install dependencies
pip install -r requirements.txt

Configuration

Edit config.yaml to customize:

server:
  host: "0.0.0.0"
  port: 8000

model:
  name: "google/medasr"
  chunk_length_s: 20
  stride_length_s: 2
  device: "auto"  # "auto", "cuda", or "cpu"

huggingface:
  token: null  # Optional: HF token for private models or rate limit increase

Usage

# Start the server
python main.py

The server will start at http://localhost:8000.

API Usage

Transcribe Audio

curl -X POST http://localhost:8000/v1/audio/transcriptions \
  -H "Content-Type: multipart/form-data" \
  -F "file=@audio.wav"

Response:

{
  "text": "transcribed text here"
}

Health Check

curl http://localhost:8000/health

Response:

{
  "status": "healthy",
  "model": "google/medasr"
}

OpenAI Client Compatible Usage

You can use this server with OpenAI client libraries by setting the base URL:

from openai import OpenAI

client = OpenAI(
    api_key="dummy",  # No actual API key needed
    base_url="http://localhost:8000/v1"
)

with open("audio.wav", "rb") as audio_file:
    transcript = client.audio.transcriptions.create(
        file=audio_file,
        model="google/medasr"
    )

print(transcript.text)

Requirements

• Python 3.10+
• transformers >= 5.0.0
• torch >= 2.0.0
• CUDA-compatible GPU (optional, for faster inference)