Try to fix confirm and interrupt

backend/requirements.txt

@@ -2,7 +2,7 @@
 # webrtc  -> SmallWebRTCTransport / SmallWebRTCConnection + aiortc
 # silero  -> 本地 VAD(判断用户说话起止),语音必备
 # openai  -> OpenAI 兼容的 LLM/STT/TTS 客户端(DeepSeek、SenseVoice、CosyVoice 都走它)
-pipecat-ai[webrtc,websocket,silero,openai,mcp]==1.5.0
+pipecat-ai[webrtc,websocket,silero,openai,mcp]==1.7.0
 Pillow>=11.1.0,<13
 
 # FastGPT 类型助手:本地 SDK(包 /api/v1/chat/completions 流式 + chatId 会话)

backend/services/pipecat/pipeline.py

@@ -107,17 +107,17 @@ ON_DEMAND_KNOWLEDGE_SYSTEM_HINT = (
 )
 
 
-async def _wait_for_interrupted_output(
-    worker: PipelineWorker,
+async def _wait_for_output_stop(
+    wait_until_stopped: Callable[[], Awaitable[None]],
     *,
-    wait_until_stopped: Callable[[], Awaitable[None]] | None = None,
+    timeout_seconds: float = 3.0,
 ) -> None:
-    """Wait until an in-band interruption crosses pipeline and playback."""
-    if not await worker.flush_pipeline(timeout=2.0):
-        raise RuntimeError("输出中断帧未能及时穿过媒体管线")
-    if wait_until_stopped is not None:
+    """Wait for the transport's real stop event after an interruption."""
     try:
-            await asyncio.wait_for(wait_until_stopped(), timeout=2.0)
+        await asyncio.wait_for(
+            wait_until_stopped(),
+            timeout=timeout_seconds,
+        )
     except TimeoutError as exc:
         raise RuntimeError("等待客户端语音停止超时") from exc
 
@@ -753,14 +753,8 @@ async def run_pipeline(
 
     async def wait_for_output_stopped() -> None:
         """Keep workflow continuation behind the interrupted output."""
-        wait_until_stopped = (
-            call_end.wait_until_silent if call_end.speaking else None
-        )
-        await _wait_for_interrupted_output(
-            worker,
-            wait_until_stopped=wait_until_stopped,
-        )
-
+        if call_end.speaking:
+            await _wait_for_output_stop(call_end.wait_until_silent)
 
     def set_system_prompt(text: str) -> None:
         """替换上下文里的系统提示(节点切换时整体替换,而非追加)。"""

backend/services/pipecat/service_factory.py

@@ -20,10 +20,12 @@ from services.pipecat.xfyun_super_tts import (
 )
 from services.pipecat.xfyun_tts import DEFAULT_XFYUN_TTS_URL, XfyunTTSService
 
-# TTS「说完」判定的空闲时长:默认 3.0s 过长(导致工作流结束节点说完后还要等约 3s
-# 才挂断,也拖慢日常轮次的交还)。设 1.0s 既能让结束语文字/音频送达,又更跟手。
-# 流式 TTS 句间音频间隔通常远小于 1s,不会把一段多句回复误判为结束。
-TTS_STOP_FRAME_TIMEOUT_S = 1.0
+# HTTP TTS may pause between response chunks. Keep Pipecat's wider default so
+# a long utterance is not marked stopped and routed onward before late chunks
+# arrive. WebSocket TTS has an explicit completion event and can use the short
+# idle fallback without delaying the end of a call.
+HTTP_TTS_STOP_FRAME_TIMEOUT_S = 3.0
+WEBSOCKET_TTS_STOP_FRAME_TIMEOUT_S = 1.0
 
 
 def config_with_resource(
@@ -158,7 +160,7 @@ def create_tts(cfg: AssistantConfig):
             volume=int(cfg.tts_values.get("volume") or 50),
             pitch=int(cfg.tts_values.get("pitch") or 50),
             push_stop_frames=True,
-            stop_frame_timeout_s=TTS_STOP_FRAME_TIMEOUT_S,
+            stop_frame_timeout_s=WEBSOCKET_TTS_STOP_FRAME_TIMEOUT_S,
         )
     if cfg.tts_interface_type not in {"openai-tts", "dashscope-tts"}:
         raise ValueError(f"不支持的 TTS 接口类型: {cfg.tts_interface_type}")
@@ -169,7 +171,7 @@ def create_tts(cfg: AssistantConfig):
     return OpenAITTSService(
         api_key=_require(cfg.tts_api_key, "TTS apiKey"),
         base_url=_require(cfg.tts_base_url, "TTS apiUrl"),
-        stop_frame_timeout_s=TTS_STOP_FRAME_TIMEOUT_S,
+        stop_frame_timeout_s=HTTP_TTS_STOP_FRAME_TIMEOUT_S,
         settings=OpenAITTSService.Settings(
             model=_require(cfg.tts_model, "TTS modelId"),
             voice=voice,

backend/tests/test_pipeline_events.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import asyncio
 import unittest
 from types import SimpleNamespace
 from unittest.mock import AsyncMock, patch
@@ -10,7 +11,7 @@ from pipecat.frames.frames import (
     OutputTransportMessageUrgentFrame,
 )
 from services.pipecat.pipeline_events import bind_cascade_pipeline_events
-from services.pipecat.pipeline import _wait_for_interrupted_output
+from services.pipecat.pipeline import _wait_for_output_stop
 
 
 class _EventSource:
@@ -81,31 +82,25 @@ class _Brain:
 
 
 class PipelineEventTest(unittest.IsolatedAsyncioTestCase):
-    async def test_interrupted_output_waits_for_flush_and_stop(self):
+    async def test_interrupted_output_waits_for_transport_stop(self):
         events = []
 
         async def wait_until_stopped():
             events.append("stopped")
 
-        worker = SimpleNamespace(
-            flush_pipeline=AsyncMock(
-                side_effect=lambda **_kwargs: events.append("flush") or True
+        await _wait_for_output_stop(wait_until_stopped)
+
+        self.assertEqual(events, ["stopped"])
+
+    async def test_interrupted_output_rejects_stop_timeout(self):
+        async def wait_forever():
+            await asyncio.Event().wait()
+
+        with self.assertRaisesRegex(RuntimeError, "语音停止超时"):
+            await _wait_for_output_stop(
+                wait_forever,
+                timeout_seconds=0.001,
             )
-        )
-
-        await _wait_for_interrupted_output(
-            worker,
-            wait_until_stopped=wait_until_stopped,
-        )
-
-        self.assertEqual(events, ["flush", "stopped"])
-        worker.flush_pipeline.assert_awaited_once_with(timeout=2.0)
-
-    async def test_interrupted_output_rejects_flush_timeout(self):
-        worker = SimpleNamespace(flush_pipeline=AsyncMock(return_value=False))
-
-        with self.assertRaisesRegex(RuntimeError, "中断帧"):
-            await _wait_for_interrupted_output(worker)
 
     async def test_greeting_keeps_playback_timestamp_until_client_ready(self):
         transport = _EventSource()

backend/tests/test_service_factory.py

@@ -0,0 +1,59 @@
+from __future__ import annotations
+
+import unittest
+from unittest.mock import patch
+
+from models import AssistantConfig
+from services.pipecat.service_factory import (
+    HTTP_TTS_STOP_FRAME_TIMEOUT_S,
+    WEBSOCKET_TTS_STOP_FRAME_TIMEOUT_S,
+    create_tts,
+)
+
+
+class TTSServiceFactoryTest(unittest.TestCase):
+    def test_http_tts_keeps_wider_audio_chunk_timeout(self):
+        config = AssistantConfig(
+            tts_interface_type="openai-tts",
+            tts_model="test-model",
+            voice="test-voice",
+            tts_api_key="test-key",
+            tts_base_url="https://tts.example.test/v1",
+        )
+
+        with patch(
+            "services.pipecat.service_factory.OpenAITTSService"
+        ) as service_type:
+            create_tts(config)
+
+        self.assertEqual(HTTP_TTS_STOP_FRAME_TIMEOUT_S, 3.0)
+        self.assertEqual(
+            service_type.call_args.kwargs["stop_frame_timeout_s"],
+            HTTP_TTS_STOP_FRAME_TIMEOUT_S,
+        )
+
+    def test_websocket_tts_keeps_short_completion_fallback(self):
+        config = AssistantConfig(
+            tts_interface_type="xfyun-tts",
+            voice="test-voice",
+            tts_secrets={
+                "appId": "test-app",
+                "apiKey": "test-key",
+                "apiSecret": "test-secret",
+            },
+        )
+
+        with patch(
+            "services.pipecat.service_factory.XfyunTTSService"
+        ) as service_type:
+            create_tts(config)
+
+        self.assertEqual(WEBSOCKET_TTS_STOP_FRAME_TIMEOUT_S, 1.0)
+        self.assertEqual(
+            service_type.call_args.kwargs["stop_frame_timeout_s"],
+            WEBSOCKET_TTS_STOP_FRAME_TIMEOUT_S,
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()