Switch to Vosk

app.py

@@ -1,15 +1,13 @@
 import os
 import json
-import wave
+import requests
 import tempfile
-import subprocess
+import shutil
 from pathlib import Path
 from fastapi import FastAPI, File, UploadFile
-from fastapi.responses import HTMLResponse, JSONResponse, StreamingResponse
+from fastapi.responses import HTMLResponse, StreamingResponse
 from fastapi.staticfiles import StaticFiles
 from fastapi.middleware.cors import CORSMiddleware
-import shutil
-import asyncio
 
 app = FastAPI()
 
@@ -22,11 +20,8 @@ app.add_middleware(
 
 app.mount("/static", StaticFiles(directory="static"), name="static")
 
-# Vosk инициализируется мгновенно, 40MB
-from vosk import Model, KaldiRecognizer
-print("🟢 Loading Vosk model...")
-model = Model("vosk-model-small-ru-0.22")
-print("✅ Vosk ready!")
+HF_TOKEN = os.getenv("HF_TOKEN")  # добавь в Settings -> Secrets
+API_URL = "https://api-inference.huggingface.co/models/openai/whisper-tiny"
 
 @app.get("/", response_class=HTMLResponse)
 async def root():
@@ -37,55 +32,27 @@ async def root():
 async def transcribe_audio(file: UploadFile = File(...)):
     async def generate():
         try:
-            yield json.dumps({"type": "status", "text": "🔄 Конвертация аудио..."}) + "\n"
+            yield json.dumps({"type": "status", "text": "🔄 Отправка на API..."}) + "\n"
             
             suffix = Path(file.filename).suffix
             with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as tmp:
                 shutil.copyfileobj(file.file, tmp)
                 tmp_path = tmp.name
             
-            # Конвертируем в WAV 16kHz моно
-            wav_path = tmp_path + ".wav"
-            subprocess.run([
-                "ffmpeg", "-i", tmp_path, 
-                "-ar", "16000", "-ac", "1", 
-                "-acodec", "pcm_s16le",
-                wav_path, "-y", "-loglevel", "quiet"
-            ])
-            
-            yield json.dumps({"type": "status", "text": "🔄 Распознавание речи..."}) + "\n"
-            
-            # Открываем WAV и распознаем
-            wf = wave.open(wav_path, "rb")
-            rec = KaldiRecognizer(model, wf.getframerate())
-            rec.SetWords(False)
-            
-            text = ""
-            while True:
-                data = wf.readframes(4000)
-                if len(data) == 0:
-                    break
-                if rec.AcceptWaveform(data):
-                    result = json.loads(rec.Result())
-                    part = result.get("text", "")
-                    if part:
-                        text += part + " "
-                        yield json.dumps({
-                            "type": "segment", 
-                            "text": text.strip()
-                        }) + "\n"
+            with open(tmp_path, "rb") as f:
+                data = f.read()
             
-            # Финальный результат
-            final = json.loads(rec.FinalResult())
-            text += final.get("text", "")
-            text = text.strip()
+            Path(tmp_path).unlink()
             
-            # Чистим файлы
-            Path(tmp_path).unlink(missing_ok=True)
-            Path(wav_path).unlink(missing_ok=True)
+            headers = {"Authorization": f"Bearer {HF_TOKEN}"}
+            response = requests.post(API_URL, headers=headers, data=data)
             
-            yield json.dumps({"type": "done", "text": text}) + "\n"
-        
+            if response.status_code == 200:
+                text = response.json().get("text", "")
+                yield json.dumps({"type": "done", "text": text}) + "\n"
+            else:
+                yield json.dumps({"type": "error", "text": f"❌ API ошибка: {response.status_code}"}) + "\n"
+                
         except Exception as e:
             yield json.dumps({"type": "error", "text": f"❌ Ошибка: {str(e)}"}) + "\n"
         finally: