Upload app.py

app.py

@@ -31,14 +31,17 @@ def generate_tts(text: str, voice: str):
     m = get_model()
     clean_voice = voice.replace("pocket-", "").replace("-", "_")
     
-    # Pocket-TTS uses generate_audio(clean_text, voice_name) or generate_audio(clean_text, voice=...)
-    try:
-        audio = m.generate_audio(clean_text, voice=clean_voice)
-    except TypeError:
-        try:
-            audio = m.generate_audio(clean_text, prompt_voice=clean_voice)
-        except TypeError:
-            audio = m.generate_audio(clean_text)
+    # 1. Get the voice state dict for the selected voice
+    if hasattr(m, "get_voice_state"):
+        voice_state = m.get_voice_state(clean_voice)
+    elif hasattr(m, "load_voice_state"):
+        voice_state = m.load_voice_state(clean_voice)
+    else:
+        # Fallback to empty dict or voice lookup
+        voice_state = m.get_state_for_voice(clean_voice) if hasattr(m, "get_state_for_voice") else {}
+
+    # 2. Generate audio with (voice_state_dict, text)
+    audio = m.generate_audio(voice_state, clean_text)
     
     if isinstance(audio, torch.Tensor):
         audio_np = audio.detach().cpu().numpy()
@@ -51,7 +54,7 @@ def generate_tts(text: str, voice: str):
 demo = gr.Interface(
     fn=generate_tts,
     inputs=[
-        gr.Textbox(label="Text to speak", value="Hello! Pocket TTS is now running on Hugging Face ZeroGPU."),
+        gr.Textbox(label="Text to speak", value="Hello! This is genuine Kyutai Pocket TTS running smoothly."),
         gr.Dropdown(choices=VOICE_LIST, value="stuart_bell", label="Voice"),
     ],
     outputs=gr.Audio(label="Generated Audio", type="numpy"),