Update idle prompt

config.json

@@ -52,7 +52,10 @@
       "没有",
       "否"
     ],
-    "user_speech_timeout_sec": 0.8
+    "user_speech_timeout_sec": 0.8,
+    "idle_prompt_timeout_sec": 8.0,
+    "idle_prompt_max_count": 1,
+    "idle_prompt_text": "我先停在这里。你可以继续说你的想法，或者让我根据刚才的内容帮你整理下一步。"
   },
   "agent": {
     "system_prompt": "You are a helpful, friendly voice assistant. Keep responses concise and natural for spoken conversation.",

config/fastgpt.state.xfyun.json

@@ -42,7 +42,10 @@
       "你好",
       "在吗"
     ],
-    "user_speech_timeout_sec": 0.2
+    "user_speech_timeout_sec": 0.8,
+    "idle_prompt_timeout_sec": 8.0,
+    "idle_prompt_max_count": 3,
+    "idle_prompt_text": "我先停在这里。你可以继续说你的想法，或者让我根据刚才的内容帮你整理下一步。"
   },
   "agent": {
     "system_prompt": "FastGPT app owns the system prompt when send_system_prompt is false.",

engine/config.py

@@ -78,6 +78,12 @@ class TurnConfig:
 
     vad: VADConfig = field(default_factory=VADConfig)
     user_speech_timeout_sec: float = 1.0
+    idle_prompt_timeout_sec: float = 0.0
+    idle_prompt_max_count: int = 1
+    idle_prompt_text: str = (
+        "我先停在这里。你可以继续说你的想法，"
+        "或者让我根据刚才的内容帮你整理下一步。"
+    )
     interruption_min_chars: int = 3
     interruption_use_interim: bool = True
     interruption_short_replies: list[str] = field(
@@ -263,6 +269,15 @@ def config_from_dict(data: dict) -> EngineConfig:
             user_speech_timeout_sec=float(
                 turn.get("user_speech_timeout_sec", TurnConfig().user_speech_timeout_sec)
             ),
+            idle_prompt_timeout_sec=float(
+                turn.get("idle_prompt_timeout_sec", TurnConfig().idle_prompt_timeout_sec)
+            ),
+            idle_prompt_max_count=int(
+                turn.get("idle_prompt_max_count", TurnConfig().idle_prompt_max_count)
+            ),
+            idle_prompt_text=str(
+                turn.get("idle_prompt_text", TurnConfig().idle_prompt_text)
+            ),
             interruption_min_chars=int(
                 turn.get("interruption_min_chars", TurnConfig().interruption_min_chars)
             ),

engine/pipeline.py

@@ -146,6 +146,7 @@ async def run_pipeline_with_serializer(
         user_params=LLMUserAggregatorParams(
             vad_analyzer=SileroVADAnalyzer(params=vad_params),
             user_turn_strategies=user_turn_strategies,
+            user_idle_timeout=config.turn.idle_prompt_timeout_sec,
         ),
     )
 
@@ -188,6 +189,7 @@ async def run_pipeline_with_serializer(
         ),
         idle_timeout_secs=config.session.inactivity_timeout_sec,
     )
+    idle_prompt_count = 0
 
     @transport.event_handler("on_client_connected")
     async def on_client_connected(_transport, _client):
@@ -216,6 +218,8 @@ async def run_pipeline_with_serializer(
 
     @user_aggregator.event_handler("on_user_turn_stopped")
     async def on_user_turn_stopped(_aggregator, _strategy, message: UserTurnStoppedMessage):
+        nonlocal idle_prompt_count
+        idle_prompt_count = 0
         logger.info(f"User: {message.content}")
         text = (message.content or "").strip()
         if not text:
@@ -231,6 +235,17 @@ async def run_pipeline_with_serializer(
             )
         )
 
+    @user_aggregator.event_handler("on_user_turn_idle")
+    async def on_user_turn_idle(_aggregator):
+        nonlocal idle_prompt_count
+        text = config.turn.idle_prompt_text.strip()
+        if not text or idle_prompt_count >= config.turn.idle_prompt_max_count:
+            return
+
+        idle_prompt_count += 1
+        logger.info("User idle prompt triggered")
+        await task.queue_frames([TTSSpeakFrame(text)])
+
     # NOTE: assistant turn started/final events are emitted by
     # ProductTextStreamProcessor, upstream of TTS, so text streams to the
     # client ahead of audio. This logger is kept for server-side visibility.