Upload app.py with huggingface_hub

app.py

@@ -59,7 +59,7 @@ if not hasattr(torchaudio, "info"):
 
 from df.enhance import enhance, init_df, load_audio, save_audio
 
-# FORCE BUILD TRIGGER: 15:35:00 Jan 20 2026
+# FORCE BUILD TRIGGER: 15:45:00 Jan 20 2026
 
 # 🛠️ Monkeypatch torchaudio.load
 try:
@@ -187,6 +187,7 @@ def core_process(request_dict):
     return {"error": f"Unknown action: {action}"}
 
 def create_wav_header(sample_rate=24000, channels=1, bit_depth=16):
+    """Returns a standard WAV header as standard BYTES"""
     header = bytearray(b'RIFF')
     header.extend((1000000000).to_bytes(4, 'little')) 
     header.extend(b'WAVEfmt ')
@@ -206,7 +207,7 @@ def create_wav_header(sample_rate=24000, channels=1, bit_depth=16):
 def gpu_tts_generator(text, lang, speaker_wav_path):
     load_models()
     try:
-        yield create_wav_header(sample_rate=24000)
+        yield bytes(create_wav_header(sample_rate=24000))
         # inference_stream is a generator
         for chunk in MODELS["tts"].synthesizer.tts_model.inference_stream(
             text,
@@ -214,7 +215,7 @@ def gpu_tts_generator(text, lang, speaker_wav_path):
             *MODELS["tts"].synthesizer.tts_model.get_conditioning_latents(audio_path=[speaker_wav_path]),
             stream_chunk_size=20
         ):
-            yield (chunk * 32767).to(torch.int16).cpu().numpy().tobytes()
+            yield bytes((chunk * 32767).to(torch.int16).cpu().numpy().tobytes())
         print("✨ [Generator Complete]")
     except Exception as e:
         print(f"❌ [Generator Error]: {e}")