Upload app.py with huggingface_hub

app.py

@@ -5,13 +5,26 @@ import base64
 import torch
 import tempfile
 import traceback
+import uvicorn
 from fastapi import FastAPI, Request
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import HTMLResponse
-import uvicorn
 
-# --- [v157] 🚀 EXTREME STABILITY ENGINE ---
-print(f"--- [v157] 📡 BOOTING ENGINE ---")
+# --- [v158] 🚀 ULTRA STABLE ENGINE ---
+# This version fixes the TorchCodec/torchaudio dependency hell on H200 ZeroGPU
+print(f"--- [v158] 📡 BOOTING ENGINE ---")
+
+# 🛠️ MONKEYPATCH torchaudio BEFORE XTTS LOADING 🛠️
+import torchaudio
+import librosa
+def stable_load(filepath, **kwargs):
+    # Redirect torchaudio.load to librosa to bypass torchcodec issues
+    # Coqui XTTS usually passes sr as a keyword or positional argument
+    target_sr = kwargs.get("sample_rate") or kwargs.get("sr") or None
+    y, sr = librosa.load(filepath, sr=target_sr)
+    return torch.from_numpy(y).unsqueeze(0), sr
+torchaudio.load = stable_load
+print("--- [v158] 🩹 TORCHAUDIO PATCH APPLIED ---")
 
 from transformers import pipeline, AutoModelForSpeechSeq2Seq, AutoProcessor
 from TTS.api import TTS
@@ -46,15 +59,15 @@ MODELS = {"stt": None, "tts": None}
 def load_tts_cpu():
     global MODELS
     if MODELS.get("tts") is None:
-        print("--- [v157] 📥 LOADING XTTS V2 (CPU) ---")
+        print("--- [v158] 📥 LOADING XTTS V2 (CPU) ---")
         MODELS["tts"] = TTS("tts_models/multilingual/multi-dataset/xtts_v2").to("cpu")
-        print("--- [v157] ✅ XTTS READY (CPU) ---")
+        print("--- [v158] ✅ XTTS READY (CPU) ---")
 
 @spaces.GPU(duration=60)
 def gpu_stt_base(temp_path, lang):
     global MODELS
     if MODELS.get("stt") is None:
-        print("--- [v157] 📥 LOADING WHISPER (Base) ON GPU ---")
+        print("--- [v158] 📥 LOADING WHISPER (Base) ON GPU ---")
         model_id = "openai/whisper-base"
         model = AutoModelForSpeechSeq2Seq.from_pretrained(model_id, torch_dtype=torch.float32).to("cuda")
         processor = AutoProcessor.from_pretrained(model_id)
@@ -65,7 +78,6 @@ def gpu_stt_base(temp_path, lang):
             feature_extractor=processor.feature_extractor,
             device="cuda"
         )
-    # Whisper Base is extremely robust and fast on H200
     res = MODELS["stt"](temp_path, generate_kwargs={"language": lang if lang and len(lang) <= 3 else None})
     return res["text"].strip()
 
@@ -74,9 +86,9 @@ async def handle_process(request: Request):
     try:
         data = await request.json()
         action = data.get("action")
-        if action == "health": return {"status": "awake", "v": "157"}
+        if action == "health": return {"status": "awake", "v": "158"}
         
-        print(f"--- [v157] 🛠️ {action} ---")
+        print(f"--- [v158] 🛠️ {action} ---")
         
         stt_text = ""
         if action in ["stt", "s2st"]:
@@ -130,20 +142,20 @@ async def handle_process(request: Request):
             return {"text": stt_text, "translated": trans_text, "audio": audio_b64}
 
     except Exception as e:
-        print(f"❌ [v157] ERROR: {traceback.format_exc()}")
+        print(f"❌ [v158] ERROR: {traceback.format_exc()}")
         return {"error": str(e)}
     finally:
-        print(f"--- [v157] ✨ DONE ({time.time()-t1:.1f}s) ---")
+        print(f"--- [v158] ✨ DONE ({time.time()-t1:.1f}s) ---")
 
 @app.post("/process")
 @app.post("/api/v1/process")
 async def api_process(request: Request): return await handle_process(request)
 
 @app.get("/health")
-def health(): return {"status": "ok", "v": "157", "gpu": torch.cuda.is_available()}
+def health(): return {"status": "ok", "v": "158", "gpu": torch.cuda.is_available()}
 
 @app.get("/", response_class=HTMLResponse)
-def root(): return "<h1>🚀 AI Engine v157 (EXTREME STABILITY)</h1>"
+def root(): return "<h1>🚀 AI Engine v158 (ULTRA STABLE)</h1>"
 
 if __name__ == "__main__":
     uvicorn.run(app, host="0.0.0.0", port=7860)