Create app.py

app.py

@@ -0,0 +1,53 @@
+# Kyutai TTS Gradio App
+
+import gradio as gr
+import numpy as np
+import torch
+from moshi.models.loaders import CheckpointInfo
+from moshi.models.tts import DEFAULT_DSM_TTS_REPO, DEFAULT_DSM_TTS_VOICE_REPO, TTSModel
+
+# Load model at startup
+checkpoint_info = CheckpointInfo.from_hf_repo(DEFAULT_DSM_TTS_REPO)
+tts_model = TTSModel.from_checkpoint_info(
+    checkpoint_info, n_q=32, temp=0.6, device=torch.device("cuda" if torch.cuda.is_available() else "cpu"),
+    dtype=torch.float16 if torch.cuda.is_available() else torch.float32
+)
+
+# Default voice
+default_voice = "expresso/ex03-ex01_happy_001_channel1_334s.wav"
+voice_path = tts_model.get_voice_path(default_voice)
+
+def tts_function(text, cfg_coef):
+    entries = tts_model.prepare_script([text], padding_between=1)
+    condition_attributes = tts_model.make_condition_attributes(
+        [voice_path], cfg_coef=cfg_coef
+    )
+
+    pcms = []
+
+    def _on_frame(frame):
+        if (frame != -1).all():
+            pcm = tts_model.mimi.decode(frame[:, 1:, :]).cpu().numpy()
+            pcms.append(np.clip(pcm[0, 0], -1, 1))
+
+    tts_model.stream(entries, condition_attributes, on_frame=_on_frame)
+
+    if pcms:
+        audio = np.concatenate(pcms)
+        return (44100, audio)
+    else:
+        return "No audio generated."
+
+# Gradio UI
+with gr.Blocks() as demo:
+    gr.Markdown("# 🎙️ Kyutai TTS - Real-Time Streaming Voice AI")
+    with gr.Row():
+        text_input = gr.Textbox(label="Enter Text", lines=5, placeholder="Type something...")
+        cfg_slider = gr.Slider(minimum=0.0, maximum=5.0, value=2.0, label="CFG Coefficient")
+    generate_btn = gr.Button("Generate Audio")
+    audio_output = gr.Audio(label="Generated Audio", type="numpy")
+
+    generate_btn.click(fn=tts_function, inputs=[text_input, cfg_slider], outputs=audio_output)
+
+if __name__ == "__main__":
+    demo.launch()