fix(voice): stop thinking continuation on silence

2026-04-16 23:56:10 +05:00
parent f14aeb39df
commit 52b2edfcd3
2 changed files with 85 additions and 2 deletions
@@ -1694,6 +1694,81 @@ def test_media_runtime_merges_thinking_continuation_into_current_utterance():
    assert merged.endswith(continuation_pcm + continuation_pcm)


+def test_media_runtime_thinking_continuation_is_not_kept_alive_by_silence():
+    runtime = AudioSocketMediaRuntime(
+        enabled=True,
+        host="127.0.0.1",
+        port=0,
+        frame_ms=20,
+        idle_timeout_seconds=2.0,
+        registration_wait_timeout_seconds=0.5,
+        min_speech_ms=40,
+        trailing_silence_ms=40,
+        max_turn_ms=2000,
+        asr_provider=_StubASRProvider(),
+        streaming_asr_provider=StreamingASRProvider(),
+        tts_provider=_StubTTSProvider(),
+        load_registration_by_media_uuid=lambda value: None,
+        mark_media_connected=lambda session_id, value: None,
+        mark_media_ended=lambda session_id, reason: None,
+        touch_media_frame=lambda session_id: None,
+        set_state=lambda session_id, state, handoff_reason, metadata: None,
+        get_pending_greeting=lambda session_id: None,
+        mark_reply_delivered=lambda session_id, text, is_greeting: None,
+        plan_reply=lambda session_id, text, metadata, kind: None,
+        process_turn=lambda session_id, transcript_text, language, barge_in, metadata: VoiceAITurnDecisionOut(
+            language=language or "ru",
+            intent="clarification",
+            reply_text="Подскажите подробнее, пожалуйста.",
+            confidence=0.8,
+            needs_handoff=False,
+            handoff_reason=None,
+            case_action="keep_open",
+            kb_refs=[],
+            summary_text="reply ready",
+            model="stub-voice",
+            latency_ms=1,
+            status="active",
+        ),
+        request_handoff=lambda session_id, customer_request_text, decision: None,
+        handle_media_error=lambda session_id, message, metadata: None,
+    )
+    base_pcm = (1000).to_bytes(2, "little", signed=True) * 160
+    continuation_pcm = (900).to_bytes(2, "little", signed=True) * 160
+    silence_frame = b"\x00\x00" * 160
+
+    async def _scenario() -> bytes:
+        actor = MediaActor(
+            registration=MediaRegistration(
+                voice_session_id="avs_media_runtime_thinking_silence_tail",
+                call_id="call_media_runtime_thinking_silence_tail",
+                interaction_id="int_media_runtime_thinking_silence_tail",
+                ai_session_id="ais_media_runtime_thinking_silence_tail",
+                language="ru",
+                media_uuid=str(uuid.uuid4()),
+            ),
+            reader=asyncio.StreamReader(),
+            writer=None,  # type: ignore[arg-type]
+            vad=EnergyVAD(frame_ms=20, min_speech_ms=40, trailing_silence_ms=40, max_turn_ms=2000),
+            frame_ms=20,
+            frame_bytes=320,
+            state="thinking",
+        )
+        merge_task = asyncio.create_task(runtime._extend_with_thinking_continuation(actor, base_pcm))
+        await runtime._handle_pcm(actor, continuation_pcm)
+        for _ in range(40):
+            await runtime._handle_pcm(actor, silence_frame)
+            await asyncio.sleep(0.02)
+        merged = await asyncio.wait_for(merge_task, timeout=0.2)
+        assert actor.thinking_continuation_pcm == bytearray()
+        return merged
+
+    merged = asyncio.run(_scenario())
+
+    assert len(merged) > len(base_pcm)
+    assert continuation_pcm in merged
+
+
 def _legacy_test_media_runtime_voice_v2_uses_streaming_sidecar_for_partial_and_final_asr():
    registrations: dict[str, MediaRegistration] = {}
    reply_starts: list[tuple[str, str | None, float]] = []