Update app.py

app.py

@@ -1,6 +1,7 @@
 import spaces
 from snac import SNAC
 import torch
+from fastapi import FastAPI, WebSocket, WebSocketDisconnect
 import gradio as gr
 import os
 from transformers import AutoModelForCausalLM, AutoTokenizer
@@ -235,7 +236,7 @@ with gr.Blocks(title="Orpheus Text-to-Speech") as demo:
         inputs=[text_input, voice, temperature, top_p, repetition_penalty, max_new_tokens],
         outputs=audio_output,
         fn=generate_speech,
-        cache_examples=True,
+        cache_examples=False,
     )
     
     # Set up event handlers
@@ -251,6 +252,34 @@ with gr.Blocks(title="Orpheus Text-to-Speech") as demo:
         outputs=[text_input, audio_output]
     )
 
-# Launch the app
+# Create FastAPI app and mount Gradio
+app = FastAPI()
+app.mount("/", demo)
+
+# WebSocket TTS endpoint\@app.websocket("/ws/tts")
+async def websocket_tts(websocket: WebSocket):
+    await websocket.accept()
+    try:
+        while True:
+            msg = await websocket.receive_text()
+            data = json.loads(msg)
+            text = data.get("text", "")
+            voice = data.get("voice", VOICES[0])
+            # Generate audio for the chunk
+            _, audio = generate_speech(text, voice, 0.7, 0.95, 1.1, 1200)
+            # Stream audio in 0.1s chunks
+            chunk_size = 2400  # 24000 Hz -> 2400 samples = 0.1s
+            for i in range(0, len(audio), chunk_size):
+                chunk = audio[i:i+chunk_size]
+                await websocket.send_bytes(chunk.tobytes())
+            await websocket.send_text("__END__")
+    except WebSocketDisconnect:
+        print("Client disconnected from /ws/tts")
+
+# Launch if run directly
+def main():
+    import uvicorn
+    uvicorn.run("app:app", host="0.0.0.0", port=7860)
+
 if __name__ == "__main__":
-    demo.queue().launch(share=False, ssr_mode=False)
+    main()