minimo reconocimiento de voz

minimal_server/RealtimeSTT/server.py

@@ -0,0 +1,23 @@
+from fastapi import FastAPI, WebSocket
+from RealtimeSTT.audio_recorder import AudioToTextRecorder
+import numpy as np
+
+app = FastAPI()
+
+recorder = AudioToTextRecorder(
+    model="tiny",
+    device="cuda",
+    compute_type="float16",
+    use_microphone=False,
+)
+
+@app.websocket("/ws/transcribe")
+async def websocket_endpoint(websocket: WebSocket):
+    await websocket.accept()
+    while True:
+        data = await websocket.receive_bytes()
+        # Convierte los bytes a numpy array (ajusta según tu formato)
+        audio = np.frombuffer(data, dtype=np.float32)
+        recorder.feed_audio(audio)
+        text = recorder.text()
+        await websocket.send_text(text)