@livekit/agents 1.0.45 → 1.0.47

package/dist/voice/audio_recognition_span.test.cjs

@@ -0,0 +1,233 @@
+"use strict";
+var import_rtc_node = require("@livekit/rtc-node");
+var import_sdk_trace_base = require("@opentelemetry/sdk-trace-base");
+var import_sdk_trace_node = require("@opentelemetry/sdk-trace-node");
+var import_vitest = require("vitest");
+var import_log = require("../log.cjs");
+var import_stt = require("../stt/stt.cjs");
+var import_telemetry = require("../telemetry/index.cjs");
+var import_vad = require("../vad.cjs");
+var import_audio_recognition = require("./audio_recognition.cjs");
+function setupInMemoryTracing() {
+  const exporter = new import_sdk_trace_base.InMemorySpanExporter();
+  const provider = new import_sdk_trace_node.NodeTracerProvider();
+  provider.addSpanProcessor(new import_sdk_trace_base.SimpleSpanProcessor(exporter));
+  provider.register();
+  (0, import_telemetry.setTracerProvider)(provider);
+  return { exporter };
+}
+function spanByName(spans, name) {
+  return spans.find((s) => s.name === name);
+}
+class FakeVADStream extends Object {
+  // We intentionally avoid extending the real VADStream (it is not exported as a value in JS output
+  // in some bundling contexts). Instead we emulate the async iterator shape used by AudioRecognition.
+  events;
+  idx = 0;
+  constructor(events) {
+    super();
+    this.events = events;
+  }
+  updateInputStream() {
+  }
+  detachInputStream() {
+  }
+  close() {
+  }
+  [Symbol.asyncIterator]() {
+    return this;
+  }
+  async next() {
+    if (this.idx >= this.events.length) {
+      return { done: true, value: void 0 };
+    }
+    const value = this.events[this.idx++];
+    return { done: false, value };
+  }
+}
+class FakeVAD extends import_vad.VAD {
+  label = "fake-vad";
+  events;
+  constructor(events) {
+    super({ updateInterval: 1 });
+    this.events = events;
+  }
+  stream() {
+    return new FakeVADStream(this.events);
+  }
+}
+const alwaysTrueTurnDetector = {
+  supportsLanguage: async () => true,
+  unlikelyThreshold: async () => void 0,
+  predictEndOfTurn: async () => 1
+};
+(0, import_vitest.describe)("AudioRecognition user_turn span parity", () => {
+  (0, import_log.initializeLogger)({ pretty: false, level: "silent" });
+  (0, import_vitest.it)("creates user_turn and parents eou_detection under it (stt mode)", async () => {
+    const { exporter } = setupInMemoryTracing();
+    const hooks = {
+      onStartOfSpeech: import_vitest.vi.fn(),
+      onVADInferenceDone: import_vitest.vi.fn(),
+      onEndOfSpeech: import_vitest.vi.fn(),
+      onInterimTranscript: import_vitest.vi.fn(),
+      onFinalTranscript: import_vitest.vi.fn(),
+      onPreemptiveGeneration: import_vitest.vi.fn(),
+      retrieveChatCtx: () => ({
+        copy() {
+          return this;
+        },
+        addMessage() {
+        },
+        toJSON() {
+          return { items: [] };
+        }
+      }),
+      onEndOfTurn: import_vitest.vi.fn(async () => true)
+    };
+    const sttEvents = [
+      { type: import_stt.SpeechEventType.START_OF_SPEECH },
+      {
+        type: import_stt.SpeechEventType.FINAL_TRANSCRIPT,
+        alternatives: [
+          {
+            language: "en",
+            text: "hello",
+            startTime: 0,
+            endTime: 0,
+            confidence: 0.9
+          }
+        ]
+      },
+      { type: import_stt.SpeechEventType.END_OF_SPEECH }
+    ];
+    const sttNode = async () => new ReadableStream({
+      start(controller) {
+        for (const ev of sttEvents) controller.enqueue(ev);
+        controller.close();
+      }
+    });
+    const ar = new import_audio_recognition.AudioRecognition({
+      recognitionHooks: hooks,
+      stt: sttNode,
+      vad: void 0,
+      turnDetector: alwaysTrueTurnDetector,
+      turnDetectionMode: "stt",
+      minEndpointingDelay: 0,
+      maxEndpointingDelay: 0,
+      sttModel: "deepgram-nova2",
+      sttProvider: "deepgram",
+      getLinkedParticipant: () => ({ sid: "p1", identity: "bob", kind: import_rtc_node.ParticipantKind.AGENT })
+    });
+    await ar.start();
+    await new Promise((r) => setTimeout(r, 20));
+    await ar.close();
+    const spans = exporter.getFinishedSpans();
+    const userTurn = spanByName(spans, "user_turn");
+    const eou = spanByName(spans, "eou_detection");
+    (0, import_vitest.expect)(userTurn, "user_turn span missing").toBeTruthy();
+    (0, import_vitest.expect)(eou, "eou_detection span missing").toBeTruthy();
+    (0, import_vitest.expect)(eou.parentSpanId).toBe(userTurn.spanContext().spanId);
+    (0, import_vitest.expect)(userTurn.attributes["lk.participant_id"]).toBe("p1");
+    (0, import_vitest.expect)(userTurn.attributes["lk.participant_identity"]).toBe("bob");
+    (0, import_vitest.expect)(userTurn.attributes["lk.participant_kind"]).toBe("AGENT");
+    (0, import_vitest.expect)(userTurn.attributes["gen_ai.request.model"]).toBe("deepgram-nova2");
+    (0, import_vitest.expect)(userTurn.attributes["gen_ai.provider.name"]).toBe("deepgram");
+    (0, import_vitest.expect)(userTurn.attributes["lk.user_transcript"]).toContain("hello");
+    (0, import_vitest.expect)(userTurn.attributes["lk.transcript_confidence"]).toBeGreaterThan(0);
+  });
+  (0, import_vitest.it)("creates user_turn from VAD startTime (vad mode) and keeps same parenting", async () => {
+    const { exporter } = setupInMemoryTracing();
+    const hooks = {
+      onStartOfSpeech: import_vitest.vi.fn(),
+      onVADInferenceDone: import_vitest.vi.fn(),
+      onEndOfSpeech: import_vitest.vi.fn(),
+      onInterimTranscript: import_vitest.vi.fn(),
+      onFinalTranscript: import_vitest.vi.fn(),
+      onPreemptiveGeneration: import_vitest.vi.fn(),
+      retrieveChatCtx: () => ({
+        copy() {
+          return this;
+        },
+        addMessage() {
+        },
+        toJSON() {
+          return { items: [] };
+        }
+      }),
+      onEndOfTurn: import_vitest.vi.fn(async () => true)
+    };
+    const now = Date.now();
+    const vadEvents = [
+      {
+        type: import_vad.VADEventType.START_OF_SPEECH,
+        samplesIndex: 0,
+        timestamp: now,
+        speechDuration: 100,
+        silenceDuration: 0,
+        frames: [],
+        probability: 0,
+        inferenceDuration: 0,
+        speaking: true,
+        rawAccumulatedSilence: 0,
+        rawAccumulatedSpeech: 0
+      },
+      {
+        type: import_vad.VADEventType.END_OF_SPEECH,
+        samplesIndex: 0,
+        timestamp: now + 200,
+        speechDuration: 100,
+        silenceDuration: 100,
+        frames: [],
+        probability: 0,
+        inferenceDuration: 0,
+        speaking: false,
+        rawAccumulatedSilence: 0,
+        rawAccumulatedSpeech: 0
+      }
+    ];
+    const sttEvents = [
+      {
+        type: import_stt.SpeechEventType.FINAL_TRANSCRIPT,
+        alternatives: [
+          {
+            language: "en",
+            text: "test",
+            startTime: 0,
+            endTime: 0,
+            confidence: 0.8
+          }
+        ]
+      }
+    ];
+    const sttNode = async () => new ReadableStream({
+      start(controller) {
+        for (const ev of sttEvents) controller.enqueue(ev);
+        controller.close();
+      }
+    });
+    const ar = new import_audio_recognition.AudioRecognition({
+      recognitionHooks: hooks,
+      stt: sttNode,
+      vad: new FakeVAD(vadEvents),
+      turnDetector: alwaysTrueTurnDetector,
+      turnDetectionMode: "vad",
+      minEndpointingDelay: 0,
+      maxEndpointingDelay: 0,
+      sttModel: "stt-model",
+      sttProvider: "stt-provider",
+      getLinkedParticipant: () => ({ sid: "p2", identity: "alice", kind: import_rtc_node.ParticipantKind.AGENT })
+    });
+    await ar.start();
+    await new Promise((r) => setTimeout(r, 20));
+    await ar.close();
+    const spans = exporter.getFinishedSpans();
+    const userTurn = spanByName(spans, "user_turn");
+    const eou = spanByName(spans, "eou_detection");
+    (0, import_vitest.expect)(userTurn).toBeTruthy();
+    (0, import_vitest.expect)(eou).toBeTruthy();
+    (0, import_vitest.expect)(eou.parentSpanId).toBe(userTurn.spanContext().spanId);
+    (0, import_vitest.expect)(hooks.onStartOfSpeech).toHaveBeenCalled();
+    (0, import_vitest.expect)(hooks.onEndOfSpeech).toHaveBeenCalled();
+  });
+});
+//# sourceMappingURL=audio_recognition_span.test.cjs.map

package/dist/voice/audio_recognition_span.test.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../src/voice/audio_recognition_span.test.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2026 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport { ParticipantKind } from '@livekit/rtc-node';\nimport { InMemorySpanExporter, SimpleSpanProcessor } from '@opentelemetry/sdk-trace-base';\nimport { NodeTracerProvider } from '@opentelemetry/sdk-trace-node';\nimport { describe, expect, it, vi } from 'vitest';\nimport { initializeLogger } from '../log.js';\nimport { type SpeechEvent, SpeechEventType } from '../stt/stt.js';\nimport { setTracerProvider } from '../telemetry/index.js';\nimport { VAD, type VADEvent, VADEventType, type VADStream } from '../vad.js';\nimport { AudioRecognition, type _TurnDetector } from './audio_recognition.js';\n\nfunction setupInMemoryTracing() {\n  const exporter = new InMemorySpanExporter();\n  const provider = new NodeTracerProvider();\n  provider.addSpanProcessor(new SimpleSpanProcessor(exporter));\n  provider.register();\n  setTracerProvider(provider);\n  return { exporter };\n}\n\nfunction spanByName(spans: any[], name: string) {\n  return spans.find((s) => s.name === name);\n}\n\nclass FakeVADStream extends (Object as unknown as { new (): VADStream }) {\n  // We intentionally avoid extending the real VADStream (it is not exported as a value in JS output\n  // in some bundling contexts). Instead we emulate the async iterator shape used by AudioRecognition.\n  private events: VADEvent[];\n  private idx = 0;\n  constructor(events: VADEvent[]) {\n    super();\n    this.events = events;\n  }\n  updateInputStream() {}\n  detachInputStream() {}\n  close() {}\n  [Symbol.asyncIterator]() {\n    return this;\n  }\n  async next(): Promise<IteratorResult<VADEvent>> {\n    if (this.idx >= this.events.length) {\n      return { done: true, value: undefined };\n    }\n    const value = this.events[this.idx++]!;\n    return { done: false, value };\n  }\n}\n\nclass FakeVAD extends VAD {\n  label = 'fake-vad';\n  private events: VADEvent[];\n  constructor(events: VADEvent[]) {\n    super({ updateInterval: 1 });\n    this.events = events;\n  }\n  stream(): any {\n    return new FakeVADStream(this.events);\n  }\n}\n\nconst alwaysTrueTurnDetector: _TurnDetector = {\n  supportsLanguage: async () => true,\n  unlikelyThreshold: async () => undefined,\n  predictEndOfTurn: async () => 1.0,\n};\n\ndescribe('AudioRecognition user_turn span parity', () => {\n  initializeLogger({ pretty: false, level: 'silent' });\n\n  it('creates user_turn and parents eou_detection under it (stt mode)', async () => {\n    const { exporter } = setupInMemoryTracing();\n\n    const hooks = {\n      onStartOfSpeech: vi.fn(),\n      onVADInferenceDone: vi.fn(),\n      onEndOfSpeech: vi.fn(),\n      onInterimTranscript: vi.fn(),\n      onFinalTranscript: vi.fn(),\n      onPreemptiveGeneration: vi.fn(),\n      retrieveChatCtx: () =>\n        ({\n          copy() {\n            return this;\n          },\n          addMessage() {},\n          toJSON() {\n            return { items: [] };\n          },\n        }) as any,\n      onEndOfTurn: vi.fn(async () => true),\n    };\n\n    const sttEvents: SpeechEvent[] = [\n      { type: SpeechEventType.START_OF_SPEECH },\n      {\n        type: SpeechEventType.FINAL_TRANSCRIPT,\n        alternatives: [\n          {\n            language: 'en',\n            text: 'hello',\n            startTime: 0,\n            endTime: 0,\n            confidence: 0.9,\n          },\n        ],\n      },\n      { type: SpeechEventType.END_OF_SPEECH },\n    ];\n\n    const sttNode = async () =>\n      new ReadableStream<SpeechEvent>({\n        start(controller) {\n          for (const ev of sttEvents) controller.enqueue(ev);\n          controller.close();\n        },\n      });\n\n    const ar = new AudioRecognition({\n      recognitionHooks: hooks as any,\n      stt: sttNode as any,\n      vad: undefined,\n      turnDetector: alwaysTrueTurnDetector,\n      turnDetectionMode: 'stt',\n      minEndpointingDelay: 0,\n      maxEndpointingDelay: 0,\n      sttModel: 'deepgram-nova2',\n      sttProvider: 'deepgram',\n      getLinkedParticipant: () => ({ sid: 'p1', identity: 'bob', kind: ParticipantKind.AGENT }),\n    });\n\n    await ar.start();\n    // allow background task to drain\n    await new Promise((r) => setTimeout(r, 20));\n    await ar.close();\n\n    const spans = exporter.getFinishedSpans();\n    const userTurn = spanByName(spans, 'user_turn');\n    const eou = spanByName(spans, 'eou_detection');\n    expect(userTurn, 'user_turn span missing').toBeTruthy();\n    expect(eou, 'eou_detection span missing').toBeTruthy();\n\n    expect(eou.parentSpanId).toBe(userTurn.spanContext().spanId);\n\n    // creation-time attributes\n    expect(userTurn.attributes['lk.participant_id']).toBe('p1');\n    expect(userTurn.attributes['lk.participant_identity']).toBe('bob');\n    expect(userTurn.attributes['lk.participant_kind']).toBe('AGENT');\n    expect(userTurn.attributes['gen_ai.request.model']).toBe('deepgram-nova2');\n    expect(userTurn.attributes['gen_ai.provider.name']).toBe('deepgram');\n\n    // end-of-turn attributes\n    expect(userTurn.attributes['lk.user_transcript']).toContain('hello');\n    expect(userTurn.attributes['lk.transcript_confidence']).toBeGreaterThan(0);\n  });\n\n  it('creates user_turn from VAD startTime (vad mode) and keeps same parenting', async () => {\n    const { exporter } = setupInMemoryTracing();\n\n    const hooks = {\n      onStartOfSpeech: vi.fn(),\n      onVADInferenceDone: vi.fn(),\n      onEndOfSpeech: vi.fn(),\n      onInterimTranscript: vi.fn(),\n      onFinalTranscript: vi.fn(),\n      onPreemptiveGeneration: vi.fn(),\n      retrieveChatCtx: () =>\n        ({\n          copy() {\n            return this;\n          },\n          addMessage() {},\n          toJSON() {\n            return { items: [] };\n          },\n        }) as any,\n      onEndOfTurn: vi.fn(async () => true),\n    };\n\n    const now = Date.now();\n    const vadEvents: VADEvent[] = [\n      {\n        type: VADEventType.START_OF_SPEECH,\n        samplesIndex: 0,\n        timestamp: now,\n        speechDuration: 100,\n        silenceDuration: 0,\n        frames: [],\n        probability: 0,\n        inferenceDuration: 0,\n        speaking: true,\n        rawAccumulatedSilence: 0,\n        rawAccumulatedSpeech: 0,\n      },\n      {\n        type: VADEventType.END_OF_SPEECH,\n        samplesIndex: 0,\n        timestamp: now + 200,\n        speechDuration: 100,\n        silenceDuration: 100,\n        frames: [],\n        probability: 0,\n        inferenceDuration: 0,\n        speaking: false,\n        rawAccumulatedSilence: 0,\n        rawAccumulatedSpeech: 0,\n      },\n    ];\n\n    const sttEvents: SpeechEvent[] = [\n      {\n        type: SpeechEventType.FINAL_TRANSCRIPT,\n        alternatives: [\n          {\n            language: 'en',\n            text: 'test',\n            startTime: 0,\n            endTime: 0,\n            confidence: 0.8,\n          },\n        ],\n      },\n    ];\n\n    const sttNode = async () =>\n      new ReadableStream<SpeechEvent>({\n        start(controller) {\n          for (const ev of sttEvents) controller.enqueue(ev);\n          controller.close();\n        },\n      });\n\n    const ar = new AudioRecognition({\n      recognitionHooks: hooks as any,\n      stt: sttNode as any,\n      vad: new FakeVAD(vadEvents) as any,\n      turnDetector: alwaysTrueTurnDetector,\n      turnDetectionMode: 'vad',\n      minEndpointingDelay: 0,\n      maxEndpointingDelay: 0,\n      sttModel: 'stt-model',\n      sttProvider: 'stt-provider',\n      getLinkedParticipant: () => ({ sid: 'p2', identity: 'alice', kind: ParticipantKind.AGENT }),\n    });\n\n    await ar.start();\n    await new Promise((r) => setTimeout(r, 20));\n    await ar.close();\n\n    const spans = exporter.getFinishedSpans();\n    const userTurn = spanByName(spans, 'user_turn');\n    const eou = spanByName(spans, 'eou_detection');\n    expect(userTurn).toBeTruthy();\n    expect(eou).toBeTruthy();\n    expect(eou.parentSpanId).toBe(userTurn.spanContext().spanId);\n\n    expect(hooks.onStartOfSpeech).toHaveBeenCalled();\n    expect(hooks.onEndOfSpeech).toHaveBeenCalled();\n  });\n});\n"],"mappings":";AAGA,sBAAgC;AAChC,4BAA0D;AAC1D,4BAAmC;AACnC,oBAAyC;AACzC,iBAAiC;AACjC,iBAAkD;AAClD,uBAAkC;AAClC,iBAAiE;AACjE,+BAAqD;AAErD,SAAS,uBAAuB;AAC9B,QAAM,WAAW,IAAI,2CAAqB;AAC1C,QAAM,WAAW,IAAI,yCAAmB;AACxC,WAAS,iBAAiB,IAAI,0CAAoB,QAAQ,CAAC;AAC3D,WAAS,SAAS;AAClB,0CAAkB,QAAQ;AAC1B,SAAO,EAAE,SAAS;AACpB;AAEA,SAAS,WAAW,OAAc,MAAc;AAC9C,SAAO,MAAM,KAAK,CAAC,MAAM,EAAE,SAAS,IAAI;AAC1C;AAEA,MAAM,sBAAuB,OAA4C;AAAA;AAAA;AAAA,EAG/D;AAAA,EACA,MAAM;AAAA,EACd,YAAY,QAAoB;AAC9B,UAAM;AACN,SAAK,SAAS;AAAA,EAChB;AAAA,EACA,oBAAoB;AAAA,EAAC;AAAA,EACrB,oBAAoB;AAAA,EAAC;AAAA,EACrB,QAAQ;AAAA,EAAC;AAAA,EACT,CAAC,OAAO,aAAa,IAAI;AACvB,WAAO;AAAA,EACT;AAAA,EACA,MAAM,OAA0C;AAC9C,QAAI,KAAK,OAAO,KAAK,OAAO,QAAQ;AAClC,aAAO,EAAE,MAAM,MAAM,OAAO,OAAU;AAAA,IACxC;AACA,UAAM,QAAQ,KAAK,OAAO,KAAK,KAAK;AACpC,WAAO,EAAE,MAAM,OAAO,MAAM;AAAA,EAC9B;AACF;AAEA,MAAM,gBAAgB,eAAI;AAAA,EACxB,QAAQ;AAAA,EACA;AAAA,EACR,YAAY,QAAoB;AAC9B,UAAM,EAAE,gBAAgB,EAAE,CAAC;AAC3B,SAAK,SAAS;AAAA,EAChB;AAAA,EACA,SAAc;AACZ,WAAO,IAAI,cAAc,KAAK,MAAM;AAAA,EACtC;AACF;AAEA,MAAM,yBAAwC;AAAA,EAC5C,kBAAkB,YAAY;AAAA,EAC9B,mBAAmB,YAAY;AAAA,EAC/B,kBAAkB,YAAY;AAChC;AAAA,IAEA,wBAAS,0CAA0C,MAAM;AACvD,mCAAiB,EAAE,QAAQ,OAAO,OAAO,SAAS,CAAC;AAEnD,wBAAG,mEAAmE,YAAY;AAChF,UAAM,EAAE,SAAS,IAAI,qBAAqB;AAE1C,UAAM,QAAQ;AAAA,MACZ,iBAAiB,iBAAG,GAAG;AAAA,MACvB,oBAAoB,iBAAG,GAAG;AAAA,MAC1B,eAAe,iBAAG,GAAG;AAAA,MACrB,qBAAqB,iBAAG,GAAG;AAAA,MAC3B,mBAAmB,iBAAG,GAAG;AAAA,MACzB,wBAAwB,iBAAG,GAAG;AAAA,MAC9B,iBAAiB,OACd;AAAA,QACC,OAAO;AACL,iBAAO;AAAA,QACT;AAAA,QACA,aAAa;AAAA,QAAC;AAAA,QACd,SAAS;AACP,iBAAO,EAAE,OAAO,CAAC,EAAE;AAAA,QACrB;AAAA,MACF;AAAA,MACF,aAAa,iBAAG,GAAG,YAAY,IAAI;AAAA,IACrC;AAEA,UAAM,YAA2B;AAAA,MAC/B,EAAE,MAAM,2BAAgB,gBAAgB;AAAA,MACxC;AAAA,QACE,MAAM,2BAAgB;AAAA,QACtB,cAAc;AAAA,UACZ;AAAA,YACE,UAAU;AAAA,YACV,MAAM;AAAA,YACN,WAAW;AAAA,YACX,SAAS;AAAA,YACT,YAAY;AAAA,UACd;AAAA,QACF;AAAA,MACF;AAAA,MACA,EAAE,MAAM,2BAAgB,cAAc;AAAA,IACxC;AAEA,UAAM,UAAU,YACd,IAAI,eAA4B;AAAA,MAC9B,MAAM,YAAY;AAChB,mBAAW,MAAM,UAAW,YAAW,QAAQ,EAAE;AACjD,mBAAW,MAAM;AAAA,MACnB;AAAA,IACF,CAAC;AAEH,UAAM,KAAK,IAAI,0CAAiB;AAAA,MAC9B,kBAAkB;AAAA,MAClB,KAAK;AAAA,MACL,KAAK;AAAA,MACL,cAAc;AAAA,MACd,mBAAmB;AAAA,MACnB,qBAAqB;AAAA,MACrB,qBAAqB;AAAA,MACrB,UAAU;AAAA,MACV,aAAa;AAAA,MACb,sBAAsB,OAAO,EAAE,KAAK,MAAM,UAAU,OAAO,MAAM,gCAAgB,MAAM;AAAA,IACzF,CAAC;AAED,UAAM,GAAG,MAAM;AAEf,UAAM,IAAI,QAAQ,CAAC,MAAM,WAAW,GAAG,EAAE,CAAC;AAC1C,UAAM,GAAG,MAAM;AAEf,UAAM,QAAQ,SAAS,iBAAiB;AACxC,UAAM,WAAW,WAAW,OAAO,WAAW;AAC9C,UAAM,MAAM,WAAW,OAAO,eAAe;AAC7C,8BAAO,UAAU,wBAAwB,EAAE,WAAW;AACtD,8BAAO,KAAK,4BAA4B,EAAE,WAAW;AAErD,8BAAO,IAAI,YAAY,EAAE,KAAK,SAAS,YAAY,EAAE,MAAM;AAG3D,8BAAO,SAAS,WAAW,mBAAmB,CAAC,EAAE,KAAK,IAAI;AAC1D,8BAAO,SAAS,WAAW,yBAAyB,CAAC,EAAE,KAAK,KAAK;AACjE,8BAAO,SAAS,WAAW,qBAAqB,CAAC,EAAE,KAAK,OAAO;AAC/D,8BAAO,SAAS,WAAW,sBAAsB,CAAC,EAAE,KAAK,gBAAgB;AACzE,8BAAO,SAAS,WAAW,sBAAsB,CAAC,EAAE,KAAK,UAAU;AAGnE,8BAAO,SAAS,WAAW,oBAAoB,CAAC,EAAE,UAAU,OAAO;AACnE,8BAAO,SAAS,WAAW,0BAA0B,CAAC,EAAE,gBAAgB,CAAC;AAAA,EAC3E,CAAC;AAED,wBAAG,4EAA4E,YAAY;AACzF,UAAM,EAAE,SAAS,IAAI,qBAAqB;AAE1C,UAAM,QAAQ;AAAA,MACZ,iBAAiB,iBAAG,GAAG;AAAA,MACvB,oBAAoB,iBAAG,GAAG;AAAA,MAC1B,eAAe,iBAAG,GAAG;AAAA,MACrB,qBAAqB,iBAAG,GAAG;AAAA,MAC3B,mBAAmB,iBAAG,GAAG;AAAA,MACzB,wBAAwB,iBAAG,GAAG;AAAA,MAC9B,iBAAiB,OACd;AAAA,QACC,OAAO;AACL,iBAAO;AAAA,QACT;AAAA,QACA,aAAa;AAAA,QAAC;AAAA,QACd,SAAS;AACP,iBAAO,EAAE,OAAO,CAAC,EAAE;AAAA,QACrB;AAAA,MACF;AAAA,MACF,aAAa,iBAAG,GAAG,YAAY,IAAI;AAAA,IACrC;AAEA,UAAM,MAAM,KAAK,IAAI;AACrB,UAAM,YAAwB;AAAA,MAC5B;AAAA,QACE,MAAM,wBAAa;AAAA,QACnB,cAAc;AAAA,QACd,WAAW;AAAA,QACX,gBAAgB;AAAA,QAChB,iBAAiB;AAAA,QACjB,QAAQ,CAAC;AAAA,QACT,aAAa;AAAA,QACb,mBAAmB;AAAA,QACnB,UAAU;AAAA,QACV,uBAAuB;AAAA,QACvB,sBAAsB;AAAA,MACxB;AAAA,MACA;AAAA,QACE,MAAM,wBAAa;AAAA,QACnB,cAAc;AAAA,QACd,WAAW,MAAM;AAAA,QACjB,gBAAgB;AAAA,QAChB,iBAAiB;AAAA,QACjB,QAAQ,CAAC;AAAA,QACT,aAAa;AAAA,QACb,mBAAmB;AAAA,QACnB,UAAU;AAAA,QACV,uBAAuB;AAAA,QACvB,sBAAsB;AAAA,MACxB;AAAA,IACF;AAEA,UAAM,YAA2B;AAAA,MAC/B;AAAA,QACE,MAAM,2BAAgB;AAAA,QACtB,cAAc;AAAA,UACZ;AAAA,YACE,UAAU;AAAA,YACV,MAAM;AAAA,YACN,WAAW;AAAA,YACX,SAAS;AAAA,YACT,YAAY;AAAA,UACd;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAEA,UAAM,UAAU,YACd,IAAI,eAA4B;AAAA,MAC9B,MAAM,YAAY;AAChB,mBAAW,MAAM,UAAW,YAAW,QAAQ,EAAE;AACjD,mBAAW,MAAM;AAAA,MACnB;AAAA,IACF,CAAC;AAEH,UAAM,KAAK,IAAI,0CAAiB;AAAA,MAC9B,kBAAkB;AAAA,MAClB,KAAK;AAAA,MACL,KAAK,IAAI,QAAQ,SAAS;AAAA,MAC1B,cAAc;AAAA,MACd,mBAAmB;AAAA,MACnB,qBAAqB;AAAA,MACrB,qBAAqB;AAAA,MACrB,UAAU;AAAA,MACV,aAAa;AAAA,MACb,sBAAsB,OAAO,EAAE,KAAK,MAAM,UAAU,SAAS,MAAM,gCAAgB,MAAM;AAAA,IAC3F,CAAC;AAED,UAAM,GAAG,MAAM;AACf,UAAM,IAAI,QAAQ,CAAC,MAAM,WAAW,GAAG,EAAE,CAAC;AAC1C,UAAM,GAAG,MAAM;AAEf,UAAM,QAAQ,SAAS,iBAAiB;AACxC,UAAM,WAAW,WAAW,OAAO,WAAW;AAC9C,UAAM,MAAM,WAAW,OAAO,eAAe;AAC7C,8BAAO,QAAQ,EAAE,WAAW;AAC5B,8BAAO,GAAG,EAAE,WAAW;AACvB,8BAAO,IAAI,YAAY,EAAE,KAAK,SAAS,YAAY,EAAE,MAAM;AAE3D,8BAAO,MAAM,eAAe,EAAE,iBAAiB;AAC/C,8BAAO,MAAM,aAAa,EAAE,iBAAiB;AAAA,EAC/C,CAAC;AACH,CAAC;","names":[]}

package/dist/voice/audio_recognition_span.test.js

@@ -0,0 +1,232 @@
+import { ParticipantKind } from "@livekit/rtc-node";
+import { InMemorySpanExporter, SimpleSpanProcessor } from "@opentelemetry/sdk-trace-base";
+import { NodeTracerProvider } from "@opentelemetry/sdk-trace-node";
+import { describe, expect, it, vi } from "vitest";
+import { initializeLogger } from "../log.js";
+import { SpeechEventType } from "../stt/stt.js";
+import { setTracerProvider } from "../telemetry/index.js";
+import { VAD, VADEventType } from "../vad.js";
+import { AudioRecognition } from "./audio_recognition.js";
+function setupInMemoryTracing() {
+  const exporter = new InMemorySpanExporter();
+  const provider = new NodeTracerProvider();
+  provider.addSpanProcessor(new SimpleSpanProcessor(exporter));
+  provider.register();
+  setTracerProvider(provider);
+  return { exporter };
+}
+function spanByName(spans, name) {
+  return spans.find((s) => s.name === name);
+}
+class FakeVADStream extends Object {
+  // We intentionally avoid extending the real VADStream (it is not exported as a value in JS output
+  // in some bundling contexts). Instead we emulate the async iterator shape used by AudioRecognition.
+  events;
+  idx = 0;
+  constructor(events) {
+    super();
+    this.events = events;
+  }
+  updateInputStream() {
+  }
+  detachInputStream() {
+  }
+  close() {
+  }
+  [Symbol.asyncIterator]() {
+    return this;
+  }
+  async next() {
+    if (this.idx >= this.events.length) {
+      return { done: true, value: void 0 };
+    }
+    const value = this.events[this.idx++];
+    return { done: false, value };
+  }
+}
+class FakeVAD extends VAD {
+  label = "fake-vad";
+  events;
+  constructor(events) {
+    super({ updateInterval: 1 });
+    this.events = events;
+  }
+  stream() {
+    return new FakeVADStream(this.events);
+  }
+}
+const alwaysTrueTurnDetector = {
+  supportsLanguage: async () => true,
+  unlikelyThreshold: async () => void 0,
+  predictEndOfTurn: async () => 1
+};
+describe("AudioRecognition user_turn span parity", () => {
+  initializeLogger({ pretty: false, level: "silent" });
+  it("creates user_turn and parents eou_detection under it (stt mode)", async () => {
+    const { exporter } = setupInMemoryTracing();
+    const hooks = {
+      onStartOfSpeech: vi.fn(),
+      onVADInferenceDone: vi.fn(),
+      onEndOfSpeech: vi.fn(),
+      onInterimTranscript: vi.fn(),
+      onFinalTranscript: vi.fn(),
+      onPreemptiveGeneration: vi.fn(),
+      retrieveChatCtx: () => ({
+        copy() {
+          return this;
+        },
+        addMessage() {
+        },
+        toJSON() {
+          return { items: [] };
+        }
+      }),
+      onEndOfTurn: vi.fn(async () => true)
+    };
+    const sttEvents = [
+      { type: SpeechEventType.START_OF_SPEECH },
+      {
+        type: SpeechEventType.FINAL_TRANSCRIPT,
+        alternatives: [
+          {
+            language: "en",
+            text: "hello",
+            startTime: 0,
+            endTime: 0,
+            confidence: 0.9
+          }
+        ]
+      },
+      { type: SpeechEventType.END_OF_SPEECH }
+    ];
+    const sttNode = async () => new ReadableStream({
+      start(controller) {
+        for (const ev of sttEvents) controller.enqueue(ev);
+        controller.close();
+      }
+    });
+    const ar = new AudioRecognition({
+      recognitionHooks: hooks,
+      stt: sttNode,
+      vad: void 0,
+      turnDetector: alwaysTrueTurnDetector,
+      turnDetectionMode: "stt",
+      minEndpointingDelay: 0,
+      maxEndpointingDelay: 0,
+      sttModel: "deepgram-nova2",
+      sttProvider: "deepgram",
+      getLinkedParticipant: () => ({ sid: "p1", identity: "bob", kind: ParticipantKind.AGENT })
+    });
+    await ar.start();
+    await new Promise((r) => setTimeout(r, 20));
+    await ar.close();
+    const spans = exporter.getFinishedSpans();
+    const userTurn = spanByName(spans, "user_turn");
+    const eou = spanByName(spans, "eou_detection");
+    expect(userTurn, "user_turn span missing").toBeTruthy();
+    expect(eou, "eou_detection span missing").toBeTruthy();
+    expect(eou.parentSpanId).toBe(userTurn.spanContext().spanId);
+    expect(userTurn.attributes["lk.participant_id"]).toBe("p1");
+    expect(userTurn.attributes["lk.participant_identity"]).toBe("bob");
+    expect(userTurn.attributes["lk.participant_kind"]).toBe("AGENT");
+    expect(userTurn.attributes["gen_ai.request.model"]).toBe("deepgram-nova2");
+    expect(userTurn.attributes["gen_ai.provider.name"]).toBe("deepgram");
+    expect(userTurn.attributes["lk.user_transcript"]).toContain("hello");
+    expect(userTurn.attributes["lk.transcript_confidence"]).toBeGreaterThan(0);
+  });
+  it("creates user_turn from VAD startTime (vad mode) and keeps same parenting", async () => {
+    const { exporter } = setupInMemoryTracing();
+    const hooks = {
+      onStartOfSpeech: vi.fn(),
+      onVADInferenceDone: vi.fn(),
+      onEndOfSpeech: vi.fn(),
+      onInterimTranscript: vi.fn(),
+      onFinalTranscript: vi.fn(),
+      onPreemptiveGeneration: vi.fn(),
+      retrieveChatCtx: () => ({
+        copy() {
+          return this;
+        },
+        addMessage() {
+        },
+        toJSON() {
+          return { items: [] };
+        }
+      }),
+      onEndOfTurn: vi.fn(async () => true)
+    };
+    const now = Date.now();
+    const vadEvents = [
+      {
+        type: VADEventType.START_OF_SPEECH,
+        samplesIndex: 0,
+        timestamp: now,
+        speechDuration: 100,
+        silenceDuration: 0,
+        frames: [],
+        probability: 0,
+        inferenceDuration: 0,
+        speaking: true,
+        rawAccumulatedSilence: 0,
+        rawAccumulatedSpeech: 0
+      },
+      {
+        type: VADEventType.END_OF_SPEECH,
+        samplesIndex: 0,
+        timestamp: now + 200,
+        speechDuration: 100,
+        silenceDuration: 100,
+        frames: [],
+        probability: 0,
+        inferenceDuration: 0,
+        speaking: false,
+        rawAccumulatedSilence: 0,
+        rawAccumulatedSpeech: 0
+      }
+    ];
+    const sttEvents = [
+      {
+        type: SpeechEventType.FINAL_TRANSCRIPT,
+        alternatives: [
+          {
+            language: "en",
+            text: "test",
+            startTime: 0,
+            endTime: 0,
+            confidence: 0.8
+          }
+        ]
+      }
+    ];
+    const sttNode = async () => new ReadableStream({
+      start(controller) {
+        for (const ev of sttEvents) controller.enqueue(ev);
+        controller.close();
+      }
+    });
+    const ar = new AudioRecognition({
+      recognitionHooks: hooks,
+      stt: sttNode,
+      vad: new FakeVAD(vadEvents),
+      turnDetector: alwaysTrueTurnDetector,
+      turnDetectionMode: "vad",
+      minEndpointingDelay: 0,
+      maxEndpointingDelay: 0,
+      sttModel: "stt-model",
+      sttProvider: "stt-provider",
+      getLinkedParticipant: () => ({ sid: "p2", identity: "alice", kind: ParticipantKind.AGENT })
+    });
+    await ar.start();
+    await new Promise((r) => setTimeout(r, 20));
+    await ar.close();
+    const spans = exporter.getFinishedSpans();
+    const userTurn = spanByName(spans, "user_turn");
+    const eou = spanByName(spans, "eou_detection");
+    expect(userTurn).toBeTruthy();
+    expect(eou).toBeTruthy();
+    expect(eou.parentSpanId).toBe(userTurn.spanContext().spanId);
+    expect(hooks.onStartOfSpeech).toHaveBeenCalled();
+    expect(hooks.onEndOfSpeech).toHaveBeenCalled();
+  });
+});
+//# sourceMappingURL=audio_recognition_span.test.js.map

package/dist/voice/audio_recognition_span.test.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../src/voice/audio_recognition_span.test.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2026 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport { ParticipantKind } from '@livekit/rtc-node';\nimport { InMemorySpanExporter, SimpleSpanProcessor } from '@opentelemetry/sdk-trace-base';\nimport { NodeTracerProvider } from '@opentelemetry/sdk-trace-node';\nimport { describe, expect, it, vi } from 'vitest';\nimport { initializeLogger } from '../log.js';\nimport { type SpeechEvent, SpeechEventType } from '../stt/stt.js';\nimport { setTracerProvider } from '../telemetry/index.js';\nimport { VAD, type VADEvent, VADEventType, type VADStream } from '../vad.js';\nimport { AudioRecognition, type _TurnDetector } from './audio_recognition.js';\n\nfunction setupInMemoryTracing() {\n  const exporter = new InMemorySpanExporter();\n  const provider = new NodeTracerProvider();\n  provider.addSpanProcessor(new SimpleSpanProcessor(exporter));\n  provider.register();\n  setTracerProvider(provider);\n  return { exporter };\n}\n\nfunction spanByName(spans: any[], name: string) {\n  return spans.find((s) => s.name === name);\n}\n\nclass FakeVADStream extends (Object as unknown as { new (): VADStream }) {\n  // We intentionally avoid extending the real VADStream (it is not exported as a value in JS output\n  // in some bundling contexts). Instead we emulate the async iterator shape used by AudioRecognition.\n  private events: VADEvent[];\n  private idx = 0;\n  constructor(events: VADEvent[]) {\n    super();\n    this.events = events;\n  }\n  updateInputStream() {}\n  detachInputStream() {}\n  close() {}\n  [Symbol.asyncIterator]() {\n    return this;\n  }\n  async next(): Promise<IteratorResult<VADEvent>> {\n    if (this.idx >= this.events.length) {\n      return { done: true, value: undefined };\n    }\n    const value = this.events[this.idx++]!;\n    return { done: false, value };\n  }\n}\n\nclass FakeVAD extends VAD {\n  label = 'fake-vad';\n  private events: VADEvent[];\n  constructor(events: VADEvent[]) {\n    super({ updateInterval: 1 });\n    this.events = events;\n  }\n  stream(): any {\n    return new FakeVADStream(this.events);\n  }\n}\n\nconst alwaysTrueTurnDetector: _TurnDetector = {\n  supportsLanguage: async () => true,\n  unlikelyThreshold: async () => undefined,\n  predictEndOfTurn: async () => 1.0,\n};\n\ndescribe('AudioRecognition user_turn span parity', () => {\n  initializeLogger({ pretty: false, level: 'silent' });\n\n  it('creates user_turn and parents eou_detection under it (stt mode)', async () => {\n    const { exporter } = setupInMemoryTracing();\n\n    const hooks = {\n      onStartOfSpeech: vi.fn(),\n      onVADInferenceDone: vi.fn(),\n      onEndOfSpeech: vi.fn(),\n      onInterimTranscript: vi.fn(),\n      onFinalTranscript: vi.fn(),\n      onPreemptiveGeneration: vi.fn(),\n      retrieveChatCtx: () =>\n        ({\n          copy() {\n            return this;\n          },\n          addMessage() {},\n          toJSON() {\n            return { items: [] };\n          },\n        }) as any,\n      onEndOfTurn: vi.fn(async () => true),\n    };\n\n    const sttEvents: SpeechEvent[] = [\n      { type: SpeechEventType.START_OF_SPEECH },\n      {\n        type: SpeechEventType.FINAL_TRANSCRIPT,\n        alternatives: [\n          {\n            language: 'en',\n            text: 'hello',\n            startTime: 0,\n            endTime: 0,\n            confidence: 0.9,\n          },\n        ],\n      },\n      { type: SpeechEventType.END_OF_SPEECH },\n    ];\n\n    const sttNode = async () =>\n      new ReadableStream<SpeechEvent>({\n        start(controller) {\n          for (const ev of sttEvents) controller.enqueue(ev);\n          controller.close();\n        },\n      });\n\n    const ar = new AudioRecognition({\n      recognitionHooks: hooks as any,\n      stt: sttNode as any,\n      vad: undefined,\n      turnDetector: alwaysTrueTurnDetector,\n      turnDetectionMode: 'stt',\n      minEndpointingDelay: 0,\n      maxEndpointingDelay: 0,\n      sttModel: 'deepgram-nova2',\n      sttProvider: 'deepgram',\n      getLinkedParticipant: () => ({ sid: 'p1', identity: 'bob', kind: ParticipantKind.AGENT }),\n    });\n\n    await ar.start();\n    // allow background task to drain\n    await new Promise((r) => setTimeout(r, 20));\n    await ar.close();\n\n    const spans = exporter.getFinishedSpans();\n    const userTurn = spanByName(spans, 'user_turn');\n    const eou = spanByName(spans, 'eou_detection');\n    expect(userTurn, 'user_turn span missing').toBeTruthy();\n    expect(eou, 'eou_detection span missing').toBeTruthy();\n\n    expect(eou.parentSpanId).toBe(userTurn.spanContext().spanId);\n\n    // creation-time attributes\n    expect(userTurn.attributes['lk.participant_id']).toBe('p1');\n    expect(userTurn.attributes['lk.participant_identity']).toBe('bob');\n    expect(userTurn.attributes['lk.participant_kind']).toBe('AGENT');\n    expect(userTurn.attributes['gen_ai.request.model']).toBe('deepgram-nova2');\n    expect(userTurn.attributes['gen_ai.provider.name']).toBe('deepgram');\n\n    // end-of-turn attributes\n    expect(userTurn.attributes['lk.user_transcript']).toContain('hello');\n    expect(userTurn.attributes['lk.transcript_confidence']).toBeGreaterThan(0);\n  });\n\n  it('creates user_turn from VAD startTime (vad mode) and keeps same parenting', async () => {\n    const { exporter } = setupInMemoryTracing();\n\n    const hooks = {\n      onStartOfSpeech: vi.fn(),\n      onVADInferenceDone: vi.fn(),\n      onEndOfSpeech: vi.fn(),\n      onInterimTranscript: vi.fn(),\n      onFinalTranscript: vi.fn(),\n      onPreemptiveGeneration: vi.fn(),\n      retrieveChatCtx: () =>\n        ({\n          copy() {\n            return this;\n          },\n          addMessage() {},\n          toJSON() {\n            return { items: [] };\n          },\n        }) as any,\n      onEndOfTurn: vi.fn(async () => true),\n    };\n\n    const now = Date.now();\n    const vadEvents: VADEvent[] = [\n      {\n        type: VADEventType.START_OF_SPEECH,\n        samplesIndex: 0,\n        timestamp: now,\n        speechDuration: 100,\n        silenceDuration: 0,\n        frames: [],\n        probability: 0,\n        inferenceDuration: 0,\n        speaking: true,\n        rawAccumulatedSilence: 0,\n        rawAccumulatedSpeech: 0,\n      },\n      {\n        type: VADEventType.END_OF_SPEECH,\n        samplesIndex: 0,\n        timestamp: now + 200,\n        speechDuration: 100,\n        silenceDuration: 100,\n        frames: [],\n        probability: 0,\n        inferenceDuration: 0,\n        speaking: false,\n        rawAccumulatedSilence: 0,\n        rawAccumulatedSpeech: 0,\n      },\n    ];\n\n    const sttEvents: SpeechEvent[] = [\n      {\n        type: SpeechEventType.FINAL_TRANSCRIPT,\n        alternatives: [\n          {\n            language: 'en',\n            text: 'test',\n            startTime: 0,\n            endTime: 0,\n            confidence: 0.8,\n          },\n        ],\n      },\n    ];\n\n    const sttNode = async () =>\n      new ReadableStream<SpeechEvent>({\n        start(controller) {\n          for (const ev of sttEvents) controller.enqueue(ev);\n          controller.close();\n        },\n      });\n\n    const ar = new AudioRecognition({\n      recognitionHooks: hooks as any,\n      stt: sttNode as any,\n      vad: new FakeVAD(vadEvents) as any,\n      turnDetector: alwaysTrueTurnDetector,\n      turnDetectionMode: 'vad',\n      minEndpointingDelay: 0,\n      maxEndpointingDelay: 0,\n      sttModel: 'stt-model',\n      sttProvider: 'stt-provider',\n      getLinkedParticipant: () => ({ sid: 'p2', identity: 'alice', kind: ParticipantKind.AGENT }),\n    });\n\n    await ar.start();\n    await new Promise((r) => setTimeout(r, 20));\n    await ar.close();\n\n    const spans = exporter.getFinishedSpans();\n    const userTurn = spanByName(spans, 'user_turn');\n    const eou = spanByName(spans, 'eou_detection');\n    expect(userTurn).toBeTruthy();\n    expect(eou).toBeTruthy();\n    expect(eou.parentSpanId).toBe(userTurn.spanContext().spanId);\n\n    expect(hooks.onStartOfSpeech).toHaveBeenCalled();\n    expect(hooks.onEndOfSpeech).toHaveBeenCalled();\n  });\n});\n"],"mappings":"AAGA,SAAS,uBAAuB;AAChC,SAAS,sBAAsB,2BAA2B;AAC1D,SAAS,0BAA0B;AACnC,SAAS,UAAU,QAAQ,IAAI,UAAU;AACzC,SAAS,wBAAwB;AACjC,SAA2B,uBAAuB;AAClD,SAAS,yBAAyB;AAClC,SAAS,KAAoB,oBAAoC;AACjE,SAAS,wBAA4C;AAErD,SAAS,uBAAuB;AAC9B,QAAM,WAAW,IAAI,qBAAqB;AAC1C,QAAM,WAAW,IAAI,mBAAmB;AACxC,WAAS,iBAAiB,IAAI,oBAAoB,QAAQ,CAAC;AAC3D,WAAS,SAAS;AAClB,oBAAkB,QAAQ;AAC1B,SAAO,EAAE,SAAS;AACpB;AAEA,SAAS,WAAW,OAAc,MAAc;AAC9C,SAAO,MAAM,KAAK,CAAC,MAAM,EAAE,SAAS,IAAI;AAC1C;AAEA,MAAM,sBAAuB,OAA4C;AAAA;AAAA;AAAA,EAG/D;AAAA,EACA,MAAM;AAAA,EACd,YAAY,QAAoB;AAC9B,UAAM;AACN,SAAK,SAAS;AAAA,EAChB;AAAA,EACA,oBAAoB;AAAA,EAAC;AAAA,EACrB,oBAAoB;AAAA,EAAC;AAAA,EACrB,QAAQ;AAAA,EAAC;AAAA,EACT,CAAC,OAAO,aAAa,IAAI;AACvB,WAAO;AAAA,EACT;AAAA,EACA,MAAM,OAA0C;AAC9C,QAAI,KAAK,OAAO,KAAK,OAAO,QAAQ;AAClC,aAAO,EAAE,MAAM,MAAM,OAAO,OAAU;AAAA,IACxC;AACA,UAAM,QAAQ,KAAK,OAAO,KAAK,KAAK;AACpC,WAAO,EAAE,MAAM,OAAO,MAAM;AAAA,EAC9B;AACF;AAEA,MAAM,gBAAgB,IAAI;AAAA,EACxB,QAAQ;AAAA,EACA;AAAA,EACR,YAAY,QAAoB;AAC9B,UAAM,EAAE,gBAAgB,EAAE,CAAC;AAC3B,SAAK,SAAS;AAAA,EAChB;AAAA,EACA,SAAc;AACZ,WAAO,IAAI,cAAc,KAAK,MAAM;AAAA,EACtC;AACF;AAEA,MAAM,yBAAwC;AAAA,EAC5C,kBAAkB,YAAY;AAAA,EAC9B,mBAAmB,YAAY;AAAA,EAC/B,kBAAkB,YAAY;AAChC;AAEA,SAAS,0CAA0C,MAAM;AACvD,mBAAiB,EAAE,QAAQ,OAAO,OAAO,SAAS,CAAC;AAEnD,KAAG,mEAAmE,YAAY;AAChF,UAAM,EAAE,SAAS,IAAI,qBAAqB;AAE1C,UAAM,QAAQ;AAAA,MACZ,iBAAiB,GAAG,GAAG;AAAA,MACvB,oBAAoB,GAAG,GAAG;AAAA,MAC1B,eAAe,GAAG,GAAG;AAAA,MACrB,qBAAqB,GAAG,GAAG;AAAA,MAC3B,mBAAmB,GAAG,GAAG;AAAA,MACzB,wBAAwB,GAAG,GAAG;AAAA,MAC9B,iBAAiB,OACd;AAAA,QACC,OAAO;AACL,iBAAO;AAAA,QACT;AAAA,QACA,aAAa;AAAA,QAAC;AAAA,QACd,SAAS;AACP,iBAAO,EAAE,OAAO,CAAC,EAAE;AAAA,QACrB;AAAA,MACF;AAAA,MACF,aAAa,GAAG,GAAG,YAAY,IAAI;AAAA,IACrC;AAEA,UAAM,YAA2B;AAAA,MAC/B,EAAE,MAAM,gBAAgB,gBAAgB;AAAA,MACxC;AAAA,QACE,MAAM,gBAAgB;AAAA,QACtB,cAAc;AAAA,UACZ;AAAA,YACE,UAAU;AAAA,YACV,MAAM;AAAA,YACN,WAAW;AAAA,YACX,SAAS;AAAA,YACT,YAAY;AAAA,UACd;AAAA,QACF;AAAA,MACF;AAAA,MACA,EAAE,MAAM,gBAAgB,cAAc;AAAA,IACxC;AAEA,UAAM,UAAU,YACd,IAAI,eAA4B;AAAA,MAC9B,MAAM,YAAY;AAChB,mBAAW,MAAM,UAAW,YAAW,QAAQ,EAAE;AACjD,mBAAW,MAAM;AAAA,MACnB;AAAA,IACF,CAAC;AAEH,UAAM,KAAK,IAAI,iBAAiB;AAAA,MAC9B,kBAAkB;AAAA,MAClB,KAAK;AAAA,MACL,KAAK;AAAA,MACL,cAAc;AAAA,MACd,mBAAmB;AAAA,MACnB,qBAAqB;AAAA,MACrB,qBAAqB;AAAA,MACrB,UAAU;AAAA,MACV,aAAa;AAAA,MACb,sBAAsB,OAAO,EAAE,KAAK,MAAM,UAAU,OAAO,MAAM,gBAAgB,MAAM;AAAA,IACzF,CAAC;AAED,UAAM,GAAG,MAAM;AAEf,UAAM,IAAI,QAAQ,CAAC,MAAM,WAAW,GAAG,EAAE,CAAC;AAC1C,UAAM,GAAG,MAAM;AAEf,UAAM,QAAQ,SAAS,iBAAiB;AACxC,UAAM,WAAW,WAAW,OAAO,WAAW;AAC9C,UAAM,MAAM,WAAW,OAAO,eAAe;AAC7C,WAAO,UAAU,wBAAwB,EAAE,WAAW;AACtD,WAAO,KAAK,4BAA4B,EAAE,WAAW;AAErD,WAAO,IAAI,YAAY,EAAE,KAAK,SAAS,YAAY,EAAE,MAAM;AAG3D,WAAO,SAAS,WAAW,mBAAmB,CAAC,EAAE,KAAK,IAAI;AAC1D,WAAO,SAAS,WAAW,yBAAyB,CAAC,EAAE,KAAK,KAAK;AACjE,WAAO,SAAS,WAAW,qBAAqB,CAAC,EAAE,KAAK,OAAO;AAC/D,WAAO,SAAS,WAAW,sBAAsB,CAAC,EAAE,KAAK,gBAAgB;AACzE,WAAO,SAAS,WAAW,sBAAsB,CAAC,EAAE,KAAK,UAAU;AAGnE,WAAO,SAAS,WAAW,oBAAoB,CAAC,EAAE,UAAU,OAAO;AACnE,WAAO,SAAS,WAAW,0BAA0B,CAAC,EAAE,gBAAgB,CAAC;AAAA,EAC3E,CAAC;AAED,KAAG,4EAA4E,YAAY;AACzF,UAAM,EAAE,SAAS,IAAI,qBAAqB;AAE1C,UAAM,QAAQ;AAAA,MACZ,iBAAiB,GAAG,GAAG;AAAA,MACvB,oBAAoB,GAAG,GAAG;AAAA,MAC1B,eAAe,GAAG,GAAG;AAAA,MACrB,qBAAqB,GAAG,GAAG;AAAA,MAC3B,mBAAmB,GAAG,GAAG;AAAA,MACzB,wBAAwB,GAAG,GAAG;AAAA,MAC9B,iBAAiB,OACd;AAAA,QACC,OAAO;AACL,iBAAO;AAAA,QACT;AAAA,QACA,aAAa;AAAA,QAAC;AAAA,QACd,SAAS;AACP,iBAAO,EAAE,OAAO,CAAC,EAAE;AAAA,QACrB;AAAA,MACF;AAAA,MACF,aAAa,GAAG,GAAG,YAAY,IAAI;AAAA,IACrC;AAEA,UAAM,MAAM,KAAK,IAAI;AACrB,UAAM,YAAwB;AAAA,MAC5B;AAAA,QACE,MAAM,aAAa;AAAA,QACnB,cAAc;AAAA,QACd,WAAW;AAAA,QACX,gBAAgB;AAAA,QAChB,iBAAiB;AAAA,QACjB,QAAQ,CAAC;AAAA,QACT,aAAa;AAAA,QACb,mBAAmB;AAAA,QACnB,UAAU;AAAA,QACV,uBAAuB;AAAA,QACvB,sBAAsB;AAAA,MACxB;AAAA,MACA;AAAA,QACE,MAAM,aAAa;AAAA,QACnB,cAAc;AAAA,QACd,WAAW,MAAM;AAAA,QACjB,gBAAgB;AAAA,QAChB,iBAAiB;AAAA,QACjB,QAAQ,CAAC;AAAA,QACT,aAAa;AAAA,QACb,mBAAmB;AAAA,QACnB,UAAU;AAAA,QACV,uBAAuB;AAAA,QACvB,sBAAsB;AAAA,MACxB;AAAA,IACF;AAEA,UAAM,YAA2B;AAAA,MAC/B;AAAA,QACE,MAAM,gBAAgB;AAAA,QACtB,cAAc;AAAA,UACZ;AAAA,YACE,UAAU;AAAA,YACV,MAAM;AAAA,YACN,WAAW;AAAA,YACX,SAAS;AAAA,YACT,YAAY;AAAA,UACd;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAEA,UAAM,UAAU,YACd,IAAI,eAA4B;AAAA,MAC9B,MAAM,YAAY;AAChB,mBAAW,MAAM,UAAW,YAAW,QAAQ,EAAE;AACjD,mBAAW,MAAM;AAAA,MACnB;AAAA,IACF,CAAC;AAEH,UAAM,KAAK,IAAI,iBAAiB;AAAA,MAC9B,kBAAkB;AAAA,MAClB,KAAK;AAAA,MACL,KAAK,IAAI,QAAQ,SAAS;AAAA,MAC1B,cAAc;AAAA,MACd,mBAAmB;AAAA,MACnB,qBAAqB;AAAA,MACrB,qBAAqB;AAAA,MACrB,UAAU;AAAA,MACV,aAAa;AAAA,MACb,sBAAsB,OAAO,EAAE,KAAK,MAAM,UAAU,SAAS,MAAM,gBAAgB,MAAM;AAAA,IAC3F,CAAC;AAED,UAAM,GAAG,MAAM;AACf,UAAM,IAAI,QAAQ,CAAC,MAAM,WAAW,GAAG,EAAE,CAAC;AAC1C,UAAM,GAAG,MAAM;AAEf,UAAM,QAAQ,SAAS,iBAAiB;AACxC,UAAM,WAAW,WAAW,OAAO,WAAW;AAC9C,UAAM,MAAM,WAAW,OAAO,eAAe;AAC7C,WAAO,QAAQ,EAAE,WAAW;AAC5B,WAAO,GAAG,EAAE,WAAW;AACvB,WAAO,IAAI,YAAY,EAAE,KAAK,SAAS,YAAY,EAAE,MAAM;AAE3D,WAAO,MAAM,eAAe,EAAE,iBAAiB;AAC/C,WAAO,MAAM,aAAa,EAAE,iBAAiB;AAAA,EAC/C,CAAC;AACH,CAAC;","names":[]}

package/dist/voice/generation.cjs

@@ -528,7 +528,7 @@ function performTextForwarding(source, controller, textOutput) {
     out
   ];
 }
-async function forwardAudio(ttsStream, audioOuput, out, signal) {
+async function forwardAudio(ttsStream, audioOutput, out, signal) {
   const reader = ttsStream.getReader();
   let resampler = null;
   const onPlaybackStarted = (ev) => {
@@ -537,8 +537,8 @@ async function forwardAudio(ttsStream, audioOuput, out, signal) {
     }
   };
   try {
-    audioOuput.on(import_io.AudioOutput.EVENT_PLAYBACK_STARTED, onPlaybackStarted);
-    audioOuput.resume();
+    audioOutput.on(import_io.AudioOutput.EVENT_PLAYBACK_STARTED, onPlaybackStarted);
+    audioOutput.resume();
     while (true) {
       if (signal == null ? void 0 : signal.aborted) {
         break;
@@ -546,29 +546,29 @@ async function forwardAudio(ttsStream, audioOuput, out, signal) {
       const { done, value: frame } = await reader.read();
       if (done) break;
       out.audio.push(frame);
-      if (!out.firstFrameFut.done && audioOuput.sampleRate && audioOuput.sampleRate !== frame.sampleRate && !resampler) {
-        resampler = new import_rtc_node.AudioResampler(frame.sampleRate, audioOuput.sampleRate, 1);
+      if (!out.firstFrameFut.done && audioOutput.sampleRate && audioOutput.sampleRate !== frame.sampleRate && !resampler) {
+        resampler = new import_rtc_node.AudioResampler(frame.sampleRate, audioOutput.sampleRate, 1);
       }
       if (resampler) {
         for (const f of resampler.push(frame)) {
-          await audioOuput.captureFrame(f);
+          await audioOutput.captureFrame(f);
         }
       } else {
-        await audioOuput.captureFrame(frame);
+        await audioOutput.captureFrame(frame);
       }
     }
     if (resampler) {
       for (const f of resampler.flush()) {
-        await audioOuput.captureFrame(f);
+        await audioOutput.captureFrame(f);
       }
     }
   } finally {
-    audioOuput.off(import_io.AudioOutput.EVENT_PLAYBACK_STARTED, onPlaybackStarted);
+    audioOutput.off(import_io.AudioOutput.EVENT_PLAYBACK_STARTED, onPlaybackStarted);
     if (!out.firstFrameFut.done) {
       out.firstFrameFut.reject(new Error("audio forwarding cancelled before playback started"));
     }
     reader == null ? void 0 : reader.releaseLock();
-    audioOuput.flush();
+    audioOutput.flush();
   }
 }
 function performAudioForwarding(ttsStream, audioOutput, controller) {
@@ -689,13 +689,6 @@ function performToolExecutions({
         },
         "Executing LLM tool call"
       );
-      const toolExecution = import_agent.asyncLocalStorage.run({ functionCall: toolCall }, async () => {
-        return await tool.execute(parsedArgs, {
-          ctx: new import_run_context.RunContext(session, speechHandle, toolCall),
-          toolCallId: toolCall.callId,
-          abortSignal: signal
-        });
-      });
       const _tracableToolExecutionImpl = async (toolExecTask, span) => {
         span.setAttribute(import_telemetry.traceTypes.ATTR_FUNCTION_TOOL_NAME, toolCall.name);
         span.setAttribute(import_telemetry.traceTypes.ATTR_FUNCTION_TOOL_ARGS, toolCall.args);
@@ -745,9 +738,36 @@ function performToolExecutions({
       const tracableToolExecution = (toolExecTask) => import_telemetry.tracer.startActiveSpan(async (span) => _tracableToolExecutionImpl(toolExecTask, span), {
         name: "function_tool"
       });
-      tasks.push(tracableToolExecution(toolExecution));
+      const toolTask = import_utils.Task.from(
+        async () => {
+          const currentTask = import_utils.Task.current();
+          if (currentTask) {
+            (0, import_agent._setActivityTaskInfo)(currentTask, {
+              speechHandle,
+              functionCall: toolCall,
+              inlineTask: true
+            });
+          }
+          const toolExecution = import_agent.functionCallStorage.run({ functionCall: toolCall }, async () => {
+            return await tool.execute(parsedArgs, {
+              ctx: new import_run_context.RunContext(session, speechHandle, toolCall),
+              toolCallId: toolCall.callId,
+              abortSignal: signal
+            });
+          });
+          await tracableToolExecution(toolExecution);
+        },
+        controller2,
+        `performToolExecution:${toolCall.name}`
+      );
+      (0, import_agent._setActivityTaskInfo)(toolTask, {
+        speechHandle,
+        functionCall: toolCall,
+        inlineTask: true
+      });
+      tasks.push(toolTask);
     }
-    await Promise.allSettled(tasks);
+    await Promise.allSettled(tasks.map((task) => task.result));
     if (toolOutput.output.length > 0) {
       logger.debug(
         {