[低延迟端点]：完成首句保留和快速结束修正，包含ACK缓冲策略、批量读帧和主说话人端点回归测试

tests/test_live_runtime.py

@@ -97,7 +97,7 @@ def make_runtime(texts: list[str], context: ConversationContext | None = None) -
     reporter = RecordingReporter()
     event_bus = PipelineEventBus()
     runtime = VoiceAssistantPipeline(
-        config=AppConfig(llm_api_key="secret", speech_provider="cloud", post_playback_drain_ms=0),
+        config=AppConfig(llm_api_key="secret", speech_provider="cloud"),
         transport=transport,
         wakeword=KeywordWakeWordProvider(),
         vad_recorder=VadRecorder(EnergyVadProvider(), min_duration_ms=40, end_silence_ms=40),
@@ -119,6 +119,7 @@ class LiveRuntimeTests(unittest.TestCase):
         summary = runtime.run(max_turns=2)
 
         self.assertEqual(summary.completed_turns, 2)
+        self.assertEqual(runtime.config.post_playback_drain_ms, 0)
         self.assertEqual(len(stt.calls), 2)
         self.assertEqual(len(llm.calls), 2)
         self.assertEqual(len(transport.played_segments), 4)

tests/test_models_config.py

@@ -73,10 +73,11 @@ class ModelsConfigTests(unittest.TestCase):
             self.assertEqual(config.wake_kws_threshold, 0.15)
             self.assertEqual(config.wake_kws_score, 1.0)
             self.assertEqual(config.wake_ack_text, "我在")
-            self.assertEqual(config.post_playback_drain_ms, 50)
+            self.assertEqual(config.post_playback_drain_ms, 0)
             self.assertEqual(config.pipeline_mode, "live_turn_based")
             self.assertEqual(config.endpoint_mode, "primary_speaker")
             self.assertEqual(config.speaker_profile_ms, 600)
+            self.assertEqual(config.speaker_profile_min_ms, 120)
             self.assertEqual(config.speaker_absent_ms, 300)
             self.assertEqual(config.speaker_similarity_threshold, 0.70)
             self.assertEqual(config.speaker_min_rms, 0.012)

tests/test_transport.py

@@ -82,6 +82,15 @@ class TransportTests(unittest.TestCase):
         self.assertEqual(frames[0].pcm, b"\x01\x00\x02\x00")
         self.assertEqual(frames[0].sample_rate, 16000)
 
+    def test_sounddevice_transport_returns_queued_frame_batch(self) -> None:
+        fake = FakeSoundDevice(callback_payloads=[b"\x01\x00", b"\x02\x00", b"\x03\x00"])
+        transport = SoundDeviceAudioTransport(sounddevice_module=fake)
+        transport.start_input(sample_rate=16000, channels=1)
+        frames = transport.read_frames(10)
+        transport.stop()
+
+        self.assertEqual([item.frame_id for item in frames], [0, 1, 2])
+
     def test_sounddevice_transport_flushes_queued_input(self) -> None:
         fake = FakeSoundDevice()
         transport = SoundDeviceAudioTransport(sounddevice_module=fake)
@@ -115,7 +124,9 @@ class FakeInputStream:
 
     def start(self) -> None:
         self.started = True
-        self.callback(b"\x01\x00\x02\x00", 2, None, "")
+        payloads = self.kwargs.get("callback_payloads") or [b"\x01\x00\x02\x00"]
+        for payload in payloads:
+            self.callback(payload, max(1, len(payload) // 2), None, "")
 
     def stop(self) -> None:
         self.started = False
@@ -140,10 +151,13 @@ class FakeOutputStream:
 
 
 class FakeSoundDevice:
-    def __init__(self) -> None:
+    def __init__(self, callback_payloads: list[bytes] | None = None) -> None:
+        self.callback_payloads = callback_payloads
         self.output_writes: list[bytes] = []
 
     def RawInputStream(self, **kwargs):
+        if self.callback_payloads is not None:
+            kwargs["callback_payloads"] = self.callback_payloads
         return FakeInputStream(**kwargs)
 
     def RawOutputStream(self, **kwargs):

tests/test_wake_vad_stt.py

@@ -116,9 +116,10 @@ class WakeVadSttTests(unittest.TestCase):
         provider.load()
         recorder = PrimarySpeakerVadRecorder(
             provider,
-            min_duration_ms=40,
+            min_duration_ms=250,
             end_silence_ms=1000,
             speaker_profile_ms=40,
+            speaker_profile_min_ms=40,
             speaker_absent_ms=40,
         )
         frames = [
@@ -142,6 +143,35 @@ class WakeVadSttTests(unittest.TestCase):
         self.assertEqual(consumed, 4)
         self.assertEqual(segment.metadata["transcript"], "你是谁")
 
+    def test_primary_speaker_endpoint_does_not_wait_for_vad_min_duration(self) -> None:
+        provider = EnergyVadProvider()
+        provider.load()
+        recorder = PrimarySpeakerVadRecorder(
+            provider,
+            min_duration_ms=1000,
+            end_silence_ms=1000,
+            speaker_profile_ms=120,
+            speaker_profile_min_ms=40,
+            speaker_absent_ms=40,
+        )
+        frames = [
+            make_frame(1, 0, speech=True, metadata={"speaker_id": "owner", "transcript": "你在做什么"}),
+            make_frame(2, 20, speech=True, metadata={"speaker_id": "owner"}),
+            make_frame(3, 40, speech=True, metadata={"speaker_id": "background"}),
+            make_frame(4, 60, speech=True, metadata={"speaker_id": "background"}),
+            make_frame(5, 80, speech=True, metadata={"speaker_id": "owner", "transcript": "第二次重复"}),
+        ]
+        segment = None
+        for item in frames:
+            result = recorder.feed(item)
+            if isinstance(result, AudioSegment):
+                segment = result
+                break
+        self.assertIsNotNone(segment)
+        assert segment is not None
+        self.assertEqual(segment.metadata["end_reason"], "primary_speaker_absent")
+        self.assertEqual(segment.metadata["transcript"], "你在做什么")
+
     def test_primary_speaker_endpoint_allows_short_pause(self) -> None:
         provider = EnergyVadProvider()
         provider.load()
@@ -150,6 +180,7 @@ class WakeVadSttTests(unittest.TestCase):
             min_duration_ms=40,
             end_silence_ms=1000,
             speaker_profile_ms=40,
+            speaker_profile_min_ms=40,
             speaker_absent_ms=60,
         )
         frames = [