Update app.py

app.py

@@ -1,30 +1,32 @@
-import gradio as gr
+# ======================
+# Setup
+# ======================
+!git clone https://github.com/kyutai-labs/delayed-streams-modeling.git
+%cd delayed-streams-modeling
+
+!pip install -q torch torchaudio gradio moshi
+
+# ======================
+# Import + Load Model
+# ======================
+import torch
 import torchaudio
-from torchao_tts import TTSModel
+from moshi.models import TTSModel
 
-# Load Kyutai TTS model
+# Load Kyutai TTS model (English + French)
 model_id = "kyutai/tts-1.6b-en_fr"
-model = TTSModel.from_pretrained(model_id)
+model = TTSModel.from_pretrained(model_id, device="cuda" if torch.cuda.is_available() else "cpu")
 
-# Function to generate speech
-def tts_fn(text, language="en"):
-    # Generate waveform
-    audio = model.generate(text, language=language)
-    # Save to temporary wav
-    torchaudio.save("output.wav", audio, 16000)
-    return "output.wav"
+# ======================
+# Simple test function
+# ======================
+def synthesize(text, lang="en", filename="out.wav"):
+    audio = model.generate(text, language=lang)
+    torchaudio.save(filename, audio.cpu(), 16000)
+    return filename
 
-# Gradio UI
-demo = gr.Interface(
-    fn=tts_fn,
-    inputs=[
-        gr.Textbox(label="Input text", placeholder="Type something to speak..."),
-        gr.Dropdown(choices=["en", "fr"], value="en", label="Language"),
-    ],
-    outputs=gr.Audio(type="filepath", label="Generated Speech"),
-    title="Kyutai TTS (1.6B EN/FR)",
-    description="Test the kyutai/tts-1.6b-en_fr model for English/French speech synthesis."
-)
+# Example: run once to check
+synthesize("Hello, this is Kyutai TTS running on Kaggle!", "en", "demo.wav")
 
-if __name__ == "__main__":
-    demo.launch()
+import IPython.display as ipd
+ipd.Audio("demo.wav")