Update app_qwen_tts.py

app_qwen_tts.py

@@ -6,6 +6,7 @@ import base64
 import requests
 from transformers import AutoTokenizer, AutoModelForCausalLM
 from sentence_transformers import SentenceTransformer
+import asyncio
 
 # =========================================================
 # Configuration
@@ -117,7 +118,6 @@ def answer_question(question):
 def tts_via_api(text: str):
     try:
         payload = {"text": text}
-        # no timeout, allow long requests
         resp = requests.post(TTS_API_URL, json=payload)
         resp.raise_for_status()
         data = resp.json()
@@ -134,26 +134,31 @@ def tts_via_api(text: str):
         return None
 
 # =========================================================
-# Gradio chat
+# Async chat
 # =========================================================
-def chat(user_message, history):
+async def chat_async(user_message, history):
     if not user_message.strip():
         return "", history
-    try:
-        # 1️⃣ User text
-        history.append((user_message, None))  # None placeholder for bot column
 
-        # 2️⃣ Generate answer text
+    # Show user message immediately
+    history.append((user_message, None))
+
+    try:
+        # 1️⃣ Generate text answer immediately
         answer_text = answer_question(user_message)
 
-        # 3️⃣ Generate audio
-        audio_path = tts_via_api(answer_text)
+        # Append answer text only (audio pending)
+        history.append((None, (answer_text, None)))
 
-        # 4️⃣ Append bot response
-        history.append((None, (answer_text, audio_path)))  # user=None, bot=(text, audio)
+        # 2️⃣ Generate audio in background
+        audio_path = await asyncio.to_thread(tts_via_api, answer_text)
+
+        # Update the last bot message with audio
+        history[-1] = (None, (answer_text, audio_path))
     except Exception as e:
         print(e)
         history.append((None, ("⚠️ Error generating response", None)))
+
     return "", history
 
 def reset_chat():
@@ -164,7 +169,8 @@ def reset_chat():
 # =========================================================
 def build_ui():
     with gr.Blocks(theme=gr.themes.Soft()) as demo:
-        gr.Markdown("# 📄 Qwen Document Assistant + TTS\nAsk a question and listen to the answer.")
+        gr.Markdown("# 📄 Qwen Document Assistant + TTS (Async Audio)")
+        gr.Markdown("Ask a question and hear the answer. Text appears immediately, audio may take a few minutes.")
 
         # Chat history: left=User, right=Bot
         chatbot = gr.Chatbot(height=450, type="tuples")
@@ -173,8 +179,8 @@ def build_ui():
         send = gr.Button("Send")
         clear = gr.Button("🧹 Clear")
 
-        send.click(chat, [msg, chatbot], [msg, chatbot])
-        msg.submit(chat, [msg, chatbot], [msg, chatbot])
+        send.click(chat_async, [msg, chatbot], [msg, chatbot])
+        msg.submit(chat_async, [msg, chatbot], [msg, chatbot])
         clear.click(reset_chat, outputs=chatbot)
 
         demo.launch(server_name="0.0.0.0", server_port=7860, share=False)