A very simplistic implementation of OpenAI Whisper's transcription API endpoint.

requirements.txt

sanic>=23.6.0

whisper.py

from sanic import Sanic, response, HTTPResponse
from sanic.request import Request

app = Sanic("TranscriptionServer")
expected_content_type = 'multipart/form-data'

@app.post("/v1/audio/transcriptions")
async def transcribe_audio(request: Request) -> HTTPResponse:
    mime_type = request.content_type.split(';')[0].strip().lower()
    if mime_type != expected_content_type:
        return response.json(body={'error': f"Content-Type must be {expected_content_type}"}, status=400)
    form = request.form
    files = request.files

    if files is None or 'file' not in files:
        return response.json(body={'error': 'No file provided'}, status=400)

    file = files.get('file')
    if file is None:
        return response.json(body={'error': 'No file selected'}, status=400)

    text = f'Mock transcription for file: {file.name} with {form}'
    print(f"Text: {text}")
    return response.json({'text': text})

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=8080, fast=True)