[真实全双工运行时]：完成run-agent-live可打断语音闭环，包含持续监听、软件回声抑制和播放取消验证

tests/test_barge_in.py

@@ -10,19 +10,21 @@ from owner_voice_pet.transport import MemoryAudioTransport
 from owner_voice_pet.vad import EnergyVadProvider
 
 
-def speech_frame(idx: int, speaker_id: str, partial: str = "等一下") -> AudioFrame:
+def speech_frame(idx: int, speaker_id: str, partial: str | None = "等一下") -> AudioFrame:
+    metadata = {
+        "duration_ms": 20,
+        "speech": True,
+        "speaker_id": speaker_id,
+    }
+    if partial is not None:
+        metadata["partial_transcript"] = partial
     return AudioFrame(
         b"\xff\x7f" * 320,
         16000,
         1,
         idx * 20,
         idx,
-        {
-            "duration_ms": 20,
-            "speech": True,
-            "speaker_id": speaker_id,
-            "partial_transcript": partial,
-        },
+        metadata,
     )
 
 
@@ -127,6 +129,74 @@ class BargeInTests(unittest.TestCase):
 
         self.assertTrue(monitor.interrupted)
 
+    def test_async_monitor_interrupts_without_realtime_partial(self) -> None:
+        gate = BargeInSpeakerGate(
+            enabled=True,
+            user_similarity_threshold=0.62,
+            assistant_reject_threshold=0.72,
+            min_rms=0.001,
+        )
+        gate.remember_user_segment(segment_for_speaker("owner"))
+        transport = MemoryAudioTransport([speech_frame(1, "owner", None), speech_frame(2, "owner", None)])
+        transport.start_input()
+        monitor = AsyncBargeInMonitor(
+            transport=transport,
+            vad_provider=EnergyVadProvider(threshold=1),
+            realtime_stt=None,
+            speaker_gate=gate,
+            assistant_profile=gate.assistant_profile(segment_for_speaker("assistant")),
+            echo_guard_ms=0,
+            min_speech_ms=40,
+            listen_interval_ms=1,
+        )
+        monitor.vad_provider.load()
+
+        monitor.start()
+        deadline = time.monotonic() + 1
+        while not monitor.interrupted and time.monotonic() < deadline:
+            time.sleep(0.005)
+        monitor.stop()
+
+        self.assertTrue(monitor.interrupted)
+        self.assertEqual(len(monitor.pending_frames()), 2)
+
+    def test_render_reference_echo_does_not_interrupt(self) -> None:
+        gate = BargeInSpeakerGate(
+            enabled=True,
+            user_similarity_threshold=0.62,
+            assistant_reject_threshold=0.72,
+            min_rms=0.001,
+        )
+        assistant = AudioSegment(b"\x01\x20\x02\x20" * 320, 16000, 1, 0, 40)
+        echo_frame = AudioFrame(
+            assistant.pcm[: 640 * 2],
+            16000,
+            1,
+            20,
+            1,
+            {"duration_ms": 40, "speech": True},
+        )
+        transport = MemoryAudioTransport([echo_frame])
+        transport.start_input()
+        monitor = AsyncBargeInMonitor(
+            transport=transport,
+            vad_provider=EnergyVadProvider(threshold=1),
+            realtime_stt=None,
+            speaker_gate=gate,
+            assistant_profile=gate.assistant_profile(assistant),
+            assistant_reference=assistant,
+            echo_guard_ms=0,
+            min_speech_ms=40,
+            listen_interval_ms=1,
+        )
+        monitor.vad_provider.load()
+
+        monitor.start()
+        time.sleep(0.05)
+        monitor.stop()
+
+        self.assertFalse(monitor.interrupted)
+
 
 if __name__ == "__main__":
     unittest.main()

tests/test_cli_acceptance.py

@@ -83,14 +83,20 @@ class CliAcceptanceTests(unittest.TestCase):
         self.assertTrue(data["success"])
         self.assertEqual(data["command"], "run-agent-live")
         self.assertEqual(data["assistant_mode"], "full_duplex_agent")
-        self.assertFalse(data["full_duplex_runtime_ready"])
+        self.assertTrue(data["full_duplex_runtime_ready"])
         self.assertEqual(data["turn_based_entry"], "run-live")
 
-    def test_run_agent_live_without_runtime_returns_explicit_not_implemented(self) -> None:
-        code, data = self.call("run-agent-live")
-        self.assertEqual(code, 1)
-        self.assertFalse(data["success"])
-        self.assertEqual(data["code"], "FULL_DUPLEX_RUNTIME_NOT_IMPLEMENTED")
+    def test_run_agent_live_once_invokes_agent_runtime(self) -> None:
+        class FakeRuntime:
+            def run_agent(self, *, once: bool = False):
+                self.once = once
+                return type("Summary", (), {"completed_turns": 1, "interrupted": False})()
+
+        fake_runtime = FakeRuntime()
+        with patch("owner_voice_pet.cli.build_live_runtime", return_value=fake_runtime):
+            code = main(["run-agent-live", "--once"])
+        self.assertEqual(code, 0)
+        self.assertTrue(fake_runtime.once)
 
     def test_security_check_command_has_no_leaks(self) -> None:
         code, data = self.call("security-check")

tests/test_live_runtime.py

@@ -261,6 +261,40 @@ def silence_frames(start_id: int, start_ms: int, count: int) -> list[AudioFrame]
 
 
 class LiveRuntimeTests(unittest.TestCase):
+    def test_agent_runtime_listens_without_wake_word(self) -> None:
+        frames = segment_frames(1, 20, partials=["直接提问", "直接提问"])
+        transport = MemoryAudioTransport(frames, flush_clears_input=False)
+        stt = QueueSttProvider(["直接提问"])
+        llm = QueueLlmProvider([["这是全双工回答。"]])
+        reporter = RecordingReporter()
+        runtime = VoiceAssistantPipeline(
+            config=AppConfig(
+                assistant_mode="full_duplex_agent",
+                llm_api_key="secret",
+                speech_provider="cloud",
+                wake_ack_text="",
+            ),
+            transport=transport,
+            wakeword=KeywordWakeWordProvider(),
+            vad_recorder=VadRecorder(EnergyVadProvider(), min_duration_ms=40, end_silence_ms=40),
+            stt=stt,
+            realtime_stt=MetadataSttProvider(),
+            llm=llm,
+            tts=SineTtsProvider(),
+            context=ConversationContext(),
+            reporter=reporter,
+            event_bus=PipelineEventBus(),
+        )
+
+        summary = runtime.run_agent(once=True)
+
+        self.assertEqual(summary.completed_turns, 1)
+        self.assertEqual(stt.calls[0].metadata["end_reason"], "silence")
+        self.assertEqual(reporter.transcripts, ["直接提问"])
+        self.assertIn("监听中：请直接说话", reporter.statuses)
+        self.assertNotIn("唤醒命中", reporter.statuses)
+        self.assertIn("恢复监听：可直接说话", reporter.statuses)
+
     def test_repeated_runtime_runs_two_turns_and_returns_to_standby(self) -> None:
         runtime, stt, llm, transport, reporter = make_runtime(["第一问", "第二问"])
         self.assertIsInstance(runtime, VoiceAssistantPipeline)