Upload app.py with huggingface_hub

app.py

@@ -10,15 +10,25 @@ from fastapi import FastAPI, Request
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import HTMLResponse
 import uvicorn
+
+# --- [v142] 🚀 REFINED HYBRID ENGINE (CPU STT + GPU TTS) ---
+print(f"--- [v142] 📡 BOOTING REFINED HYBRID ---")
+
+# Top-level imports for stability
 from transformers import pipeline, AutoModelForSpeechSeq2Seq, AutoProcessor
 from TTS.api import TTS
-
-# --- [v141] 🚀 HYBRID STABLE ENGINE (CPU STT + GPU TTS) ---
-print(f"--- [v141] 📡 BOOTING HYBRID ENGINE ---")
+from deep_translator import GoogleTranslator
+try:
+    import chatterbox_utils
+    HAS_CHATTERBOX = True
+except ImportError:
+    HAS_CHATTERBOX = False
 
 try:
     import spaces
+    HAS_SPACES = True
 except ImportError:
+    HAS_SPACES = False
     class spaces:
         @staticmethod
         def GPU(duration=60, f=None):
@@ -38,12 +48,11 @@ def load_cpu_stt():
     """Loads Whisper on CPU for maximum stability."""
     global MODELS
     if MODELS.get("stt") is None:
-        print("--- [v141] 📥 LOADING WHISPER ON CPU ---")
+        print("--- [v142] 📥 LOADING WHISPER ON CPU ---")
         model_id = "openai/whisper-large-v3-turbo"
-        # CPU loading (fp32)
         model = AutoModelForSpeechSeq2Seq.from_pretrained(
             model_id, torch_dtype=torch.float32, low_cpu_mem_usage=True
-        )
+        ) # CPU handles FP32 best
         processor = AutoProcessor.from_pretrained(model_id)
         MODELS["stt"] = pipeline(
             "automatic-speech-recognition",
@@ -52,18 +61,18 @@ def load_cpu_stt():
             feature_extractor=processor.feature_extractor,
             device="cpu"
         )
-        print("--- [v141] ✅ WHISPER LOADED (CPU) ---")
+        print("--- [v142] ✅ WHISPER LOADED (CPU) ---")
 
-@spaces.GPU(duration=90)
+@spaces.GPU(duration=120)
 def load_gpu_tts():
     """Loads XTTS on GPU for maximum speed."""
     global MODELS
     if MODELS.get("tts") is None:
-        print("--- [v141] 📥 LOADING XTTS ON GPU ---")
+        print("--- [v142] 📥 LOADING XTTS ON GPU ---")
         MODELS["tts"] = TTS("tts_models/multilingual/multi-dataset/xtts_v2").to("cuda")
-        print("--- [v141] ✅ XTTS LOADED (GPU) ---")
+        print("--- [v142] ✅ XTTS LOADED (GPU) ---")
 
-def stt_process(audio_b64, lang):
+def stt_process_internal(audio_b64, lang):
     load_cpu_stt()
     audio_bytes = base64.b64decode(audio_b64)
     with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as f:
@@ -74,20 +83,27 @@ def stt_process(audio_b64, lang):
     finally:
         if os.path.exists(temp_path): os.unlink(temp_path)
 
-@spaces.GPU(duration=90)
-def tts_process(text, target_lang, speaker_wav_b64=None):
+@spaces.GPU(duration=120)
+def tts_process_internal(text, target_lang, speaker_wav_b64=None):
     load_gpu_tts()
     XTTS_MAP = {"en": "en", "de": "de", "fr": "fr", "es": "es", "it": "it", "pl": "pl", "pt": "pt", "tr": "tr", "ru": "ru", "nl": "nl", "cs": "cs", "ar": "ar", "hu": "hu", "ko": "ko", "hi": "hi", "zh": "zh-cn"}
     clean_lang = target_lang.split('-')[0].lower()
     mapped_lang = XTTS_MAP.get(clean_lang) or ("zh-cn" if clean_lang == "zh" else None)
     
-    if not mapped_lang: return {"error": f"Language {clean_lang} not supported."}
+    if not mapped_lang: 
+        if HAS_CHATTERBOX:
+            print(f"--- [v142] 📦 USING CHATTERBOX FOR {clean_lang} ---")
+            audio_bytes = chatterbox_utils.run_chatterbox_inference(text, clean_lang)
+            return base64.b64encode(audio_bytes).decode()
+        return {"error": f"Language {clean_lang} not supported."}
     
     speaker_wav_path = "default_speaker.wav"
     if speaker_wav_b64:
         sb = base64.b64decode(speaker_wav_b64)
         with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as f:
             f.write(sb); speaker_wav_path = f.name
+    elif not os.path.exists(speaker_wav_path):
+        speaker_wav_path = None
 
     try:
         with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as out_f: out_p = out_f.name
@@ -103,46 +119,46 @@ async def api_process(request: Request):
     try:
         data = await request.json()
         action = data.get("action")
-        if action == "health": return {"status": "awake", "v": "141"}
+        if action == "health": return {"status": "awake", "v": "142", "gpu": HAS_SPACES}
         
-        print(f"--- [v141] 🛠️ HYBRID ENGINE: {action} ---")
+        print(f"--- [v142] 🛠️ HYBRID ENGINE: {action} ---")
         t1 = time.time()
         
-        # 🎙️ STT (CPU)
+        # 🎙️ STT (CPU Segment)
         stt_text = None
         if action in ["stt", "s2st"]:
-            stt_text = stt_process(data.get("file"), data.get("lang"))
+            stt_text = stt_process_internal(data.get("file"), data.get("lang"))
             if action == "stt": return {"text": stt_text}
 
-        # 🔊 TTS (GPU)
+        # 🔊 TTS (GPU Segment)
         if action in ["tts", "s2st"]:
             text = (data.get("text") if action == "tts" else stt_text).strip()
             trans_text = text
+            
             if action == "s2st":
-                from deep_translator import GoogleTranslator
                 target = data.get("target_lang") or "en"
                 trans_text = GoogleTranslator(source='auto', target=target).translate(stt_text)
                 text = trans_text
 
             if len(text) < 2: return {"audio": ""} if action == "tts" else {"text": stt_text, "translated": "", "audio": ""}
             
-            audio_res = tts_process(text, (data.get("lang") if action == "tts" else data.get("target_lang")), data.get("speaker_wav"))
+            audio_res = tts_process_internal(text, (data.get("lang") if action == "tts" else target), data.get("speaker_wav"))
             if isinstance(audio_res, dict) and "error" in audio_res: return audio_res
             
             if action == "tts": return {"audio": audio_res}
             return {"text": stt_text, "translated": trans_text, "audio": audio_res}
 
     except Exception as e:
-        print(f"❌ [v141] ERROR: {traceback.format_exc()}")
+        print(f"❌ [v142] ERROR: {traceback.format_exc()}")
         return {"error": str(e)}
     finally:
-        print(f"--- [v141] ✨ DONE ---")
+        print(f"--- [v142] ✨ DONE ({time.time()-t1:.1f}s) ---")
 
 @app.get("/health")
-def health(): return {"status": "ok", "v": "141", "mode": "HYBRID"}
+def health(): return {"status": "ok", "v": "142", "mode": "HYBRID", "gpu": HAS_SPACES}
 
 @app.get("/", response_class=HTMLResponse)
-def root(): return "<html><body><h1>🚀 AI Engine v141 (HYBRID STABLE)</h1></body></html>"
+def root(): return "<html><body><h1>🚀 AI Engine v142 (PRODUCTION HYBRID)</h1></body></html>"
 
 if __name__ == "__main__":
     uvicorn.run(app, host="0.0.0.0", port=7860)