server.py

import numpy as np
from fastapi import FastAPI, WebSocket
from faster_whisper import WhisperModel
import uvicorn
"""The core of the real-time processing. It receives the audio stream sent by clients,
 segments it and transmits it to a speech recognition model (STT)."""

app = FastAPI()

# --- Charger le modèle UNE seule fois ---
model = WhisperModel("small", compute_type="int8")
SAMPLE_RATE = 16000
CHUNK_SECONDS = 1.0
BUFFER_SIZE = int(SAMPLE_RATE * CHUNK_SECONDS)

@app.websocket("/ws/transcribe")
async def websocket_transcribe(ws: WebSocket):
    await ws.accept()
    print("Client connected")

    audio_buffer = np.array([], dtype=np.float32)

    try:
        while True:
            try:
                data = await ws.receive_bytes()
            except:
                print("Client disconnected")
                break

            chunk = np.frombuffer(data, np.int16).astype(np.float32) / 32768.0
            audio_buffer = np.concatenate((audio_buffer, chunk))

            # Transcrire seulement si on a assez
            if len(audio_buffer) >= BUFFER_SIZE:
                segments, _ = model.transcribe(audio_buffer, language="en")
                text = " ".join([seg.text for seg in segments])
                await ws.send_text(text)

                # Rolling buffer : garder dernier 0.5s pour contexte
                audio_buffer = audio_buffer[-int(SAMPLE_RATE * 0.5):]

    except Exception as e:
        print("WebSocket closed", e)

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8000)