feat(speech): enable local inference instead of calling api

app/.env.example

@@ -4,4 +4,5 @@ MQTT_PORT=1883
 ADAFRUIT_USER=tuankiet0303
 ADAFRUIT_KEY=
 MONGO_USER=root
-MONGO_PASSWORD=example
+MONGO_PASSWORD=example
+HUGGINGFACE_KEY=

app/app.py

@@ -11,7 +11,7 @@
 from flask import Flask, Response, render_template, request
 from flask_cors import CORS
 from flask_socketio import SocketIO, emit
-from inference import query
+from local_inference import query
 from myMqtt import *
 from pymongo import MongoClient
 from werkzeug.utils import secure_filename

app/inference.py

@@ -1,8 +1,8 @@
 import asyncio
-
+import os
 from aiohttp import ClientSession
 
-headers = {"Authorization": f"Bearer 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"}
+headers = {"Authorization": f"Bearer {os.environ.get("HUGGINGFACE_KEY")}"}
 API_URL = "https://api-inference.huggingface.co/models/vinai/PhoWhisper-base"
 
 

app/local_inference.py

@@ -0,0 +1,18 @@
+import torch
+from transformers import pipeline
+import asyncio
+
+device = "cuda:0" if torch.cuda.is_available() else "cpu"
+
+model_id = "vinai/PhoWhisper-base"
+
+transcriber = pipeline("automatic-speech-recognition", model="vinai/PhoWhisper-base",device=device)
+async def query(filename):
+    return transcriber(filename)
+
+async def main():
+    output = await query("uploads/audio.flac")
+    print(output)
+
+if __name__ == "__main__":
+    asyncio.run(main())