@livekit/agents 1.0.44 → 1.0.46

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/io.cjs

@@ -30,7 +30,7 @@ __export(io_exports, {
 module.exports = __toCommonJS(io_exports);
 var import_node_events = require("node:events");
 var import_log = require("../log.cjs");
-var import_deferred_stream = require("../stream/deferred_stream.cjs");
+var import_multi_input_stream = require("../stream/multi_input_stream.cjs");
 var import_utils = require("../utils.cjs");
 const TIMED_STRING_SYMBOL = Symbol.for("lk.TimedString");
 function createTimedString(opts) {
@@ -47,9 +47,12 @@ function isTimedString(value) {
   return typeof value === "object" && value !== null && TIMED_STRING_SYMBOL in value && value[TIMED_STRING_SYMBOL] === true;
 }
 class AudioInput {
-  deferredStream = new import_deferred_stream.DeferredReadableStream();
+  multiStream = new import_multi_input_stream.MultiInputStream();
   get stream() {
-    return this.deferredStream.stream;
+    return this.multiStream.stream;
+  }
+  async close() {
+    await this.multiStream.close();
   }
   onAttached() {
   }

package/dist/voice/io.cjs.map

@@ -1 +1 @@
-{"version":3,"sources":["../../src/voice/io.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2025 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport type { AudioFrame } from '@livekit/rtc-node';\nimport { EventEmitter } from 'node:events';\nimport type { ReadableStream } from 'node:stream/web';\nimport type { ChatContext } from '../llm/chat_context.js';\nimport type { ChatChunk } from '../llm/llm.js';\nimport type { ToolContext } from '../llm/tool_context.js';\nimport { log } from '../log.js';\nimport { DeferredReadableStream } from '../stream/deferred_stream.js';\nimport type { SpeechEvent } from '../stt/stt.js';\nimport { Future } from '../utils.js';\nimport type { ModelSettings } from './agent.js';\n\nexport type STTNode = (\n  audio: ReadableStream<AudioFrame>,\n  modelSettings: ModelSettings,\n) => Promise<ReadableStream<SpeechEvent | string> | null>;\n\nexport type LLMNode = (\n  chatCtx: ChatContext,\n  toolCtx: ToolContext,\n  modelSettings: ModelSettings,\n) => Promise<ReadableStream<ChatChunk | string> | null>;\n\nexport type TTSNode = (\n  text: ReadableStream<string>,\n  modelSettings: ModelSettings,\n) => Promise<ReadableStream<AudioFrame> | null>;\n\n/**\n * Symbol used to identify TimedString objects.\n */\nexport const TIMED_STRING_SYMBOL = Symbol.for('lk.TimedString');\n\n/**\n * A string with optional start and end timestamps for word-level alignment.\n */\nexport interface TimedString {\n  readonly [TIMED_STRING_SYMBOL]: true;\n  text: string;\n  startTime?: number; // seconds\n  endTime?: number; // seconds\n  confidence?: number;\n  startTimeOffset?: number;\n}\n\n/**\n * Factory function to create a TimedString object.\n */\nexport function createTimedString(opts: {\n  text: string;\n  startTime?: number;\n  endTime?: number;\n  confidence?: number;\n  startTimeOffset?: number;\n}): TimedString {\n  return {\n    [TIMED_STRING_SYMBOL]: true,\n    text: opts.text,\n    startTime: opts.startTime,\n    endTime: opts.endTime,\n    confidence: opts.confidence,\n    startTimeOffset: opts.startTimeOffset,\n  };\n}\n\n/**\n * Type guard to check if a value is a TimedString.\n */\nexport function isTimedString(value: unknown): value is TimedString {\n  return (\n    typeof value === 'object' &&\n    value !== null &&\n    TIMED_STRING_SYMBOL in value &&\n    (value as TimedString)[TIMED_STRING_SYMBOL] === true\n  );\n}\n\nexport interface AudioOutputCapabilities {\n  /** Whether this output supports pause/resume functionality */\n  pause: boolean;\n}\n\nexport abstract class AudioInput {\n  protected deferredStream: DeferredReadableStream<AudioFrame> =\n    new DeferredReadableStream<AudioFrame>();\n\n  get stream(): ReadableStream<AudioFrame> {\n    return this.deferredStream.stream;\n  }\n\n  onAttached(): void {}\n\n  onDetached(): void {}\n}\n\nexport abstract class AudioOutput extends EventEmitter {\n  static readonly EVENT_PLAYBACK_STARTED = 'playbackStarted';\n  static readonly EVENT_PLAYBACK_FINISHED = 'playbackFinished';\n\n  private playbackFinishedFuture: Future<void> = new Future();\n  private _capturing: boolean = false;\n  private playbackFinishedCount: number = 0;\n  private playbackSegmentsCount: number = 0;\n  private lastPlaybackEvent: PlaybackFinishedEvent = {\n    playbackPosition: 0,\n    interrupted: false,\n  };\n  protected logger = log();\n  protected readonly capabilities: AudioOutputCapabilities;\n\n  constructor(\n    public sampleRate?: number,\n    protected readonly nextInChain?: AudioOutput,\n    capabilities: AudioOutputCapabilities = { pause: false },\n  ) {\n    super();\n    this.capabilities = capabilities;\n\n    if (this.nextInChain) {\n      this.nextInChain.on(AudioOutput.EVENT_PLAYBACK_STARTED, (ev: PlaybackStartedEvent) =>\n        this.onPlaybackStarted(ev.createdAt),\n      );\n      this.nextInChain.on(AudioOutput.EVENT_PLAYBACK_FINISHED, (ev: PlaybackFinishedEvent) =>\n        this.onPlaybackFinished(ev),\n      );\n    }\n  }\n\n  /**\n   * Whether this output and all outputs in the chain support pause/resume.\n   */\n  get canPause(): boolean {\n    return this.capabilities.pause && (this.nextInChain?.canPause ?? true);\n  }\n\n  /**\n   * Capture an audio frame for playback, frames can be pushed faster than real-time\n   */\n  async captureFrame(_frame: AudioFrame): Promise<void> {\n    if (!this._capturing) {\n      this._capturing = true;\n      this.playbackSegmentsCount++;\n    }\n  }\n\n  /**\n   * Wait for the past audio segments to finish playing out.\n   *\n   * @returns The event that was emitted when the audio finished playing out (only the last segment information)\n   */\n  async waitForPlayout(): Promise<PlaybackFinishedEvent> {\n    const target = this.playbackSegmentsCount;\n\n    while (this.playbackFinishedCount < target) {\n      await this.playbackFinishedFuture.await;\n      this.playbackFinishedFuture = new Future();\n    }\n\n    return this.lastPlaybackEvent;\n  }\n\n  /**\n   * Called when playback actually starts (first frame is sent to output).\n   * Developers building audio sinks should call this when the first frame is captured.\n   */\n  onPlaybackStarted(createdAt: number): void {\n    this.emit(AudioOutput.EVENT_PLAYBACK_STARTED, { createdAt } as PlaybackStartedEvent);\n  }\n\n  /**\n   * Developers building audio sinks must call this method when a playback/segment is finished.\n   * Segments are segmented by calls to flush() or clearBuffer()\n   */\n  onPlaybackFinished(options: PlaybackFinishedEvent) {\n    if (this.playbackFinishedCount >= this.playbackSegmentsCount) {\n      this.logger.warn('playback_finished called more times than playback segments were captured');\n      return;\n    }\n\n    this.lastPlaybackEvent = options;\n    this.playbackFinishedCount++;\n    this.playbackFinishedFuture.resolve();\n    this.emit(AudioOutput.EVENT_PLAYBACK_FINISHED, options);\n  }\n\n  flush(): void {\n    this._capturing = false;\n  }\n\n  /**\n   * Clear the buffer, stopping playback immediately\n   */\n  abstract clearBuffer(): void;\n\n  onAttached(): void {\n    if (this.nextInChain) {\n      this.nextInChain.onAttached();\n    }\n  }\n\n  onDetached(): void {\n    if (this.nextInChain) {\n      this.nextInChain.onDetached();\n    }\n  }\n\n  /**\n   * Pause the audio playback\n   */\n  pause(): void {\n    if (this.nextInChain) {\n      this.nextInChain.pause();\n    }\n  }\n\n  /**\n   * Resume the audio playback\n   */\n  resume(): void {\n    if (this.nextInChain) {\n      this.nextInChain.resume();\n    }\n  }\n}\n\nexport interface PlaybackFinishedEvent {\n  /** How much of the audio was played back, in seconds */\n  playbackPosition: number;\n  /** True if playback was interrupted (clearBuffer() was called) */\n  interrupted: boolean;\n  /**\n   * Transcript synced with playback; may be partial if the audio was interrupted.\n   * When undefined, the transcript is not synchronized with the playback.\n   */\n  synchronizedTranscript?: string;\n}\n\nexport interface PlaybackStartedEvent {\n  /** The timestamp (Date.now()) when the playback started */\n  createdAt: number;\n}\n\nexport abstract class TextOutput {\n  constructor(protected readonly nextInChain?: TextOutput) {}\n\n  abstract captureText(text: string | TimedString): Promise<void>;\n\n  /**\n   * Mark the current text segment as complete (e.g LLM generation is complete)\n   */\n  abstract flush(): void;\n\n  onAttached(): void {\n    if (this.nextInChain) {\n      this.nextInChain.onAttached();\n    }\n  }\n\n  onDetached(): void {\n    if (this.nextInChain) {\n      this.nextInChain.onDetached();\n    }\n  }\n}\n\nexport class AgentInput {\n  private _audioStream: AudioInput | null = null;\n  // enabled by default\n  private _audioEnabled: boolean = true;\n\n  constructor(private readonly audioChanged: () => void) {}\n\n  setAudioEnabled(enable: boolean): void {\n    if (enable === this._audioEnabled) {\n      return;\n    }\n\n    this._audioEnabled = enable;\n\n    if (!this._audioStream) {\n      return;\n    }\n\n    if (enable) {\n      this._audioStream.onAttached();\n    } else {\n      this._audioStream.onDetached();\n    }\n  }\n\n  get audioEnabled(): boolean {\n    return this._audioEnabled;\n  }\n\n  get audio(): AudioInput | null {\n    return this._audioStream;\n  }\n\n  set audio(stream: AudioInput | null) {\n    this._audioStream = stream;\n    this.audioChanged();\n  }\n}\n\nexport class AgentOutput {\n  private _audioSink: AudioOutput | null = null;\n  private _transcriptionSink: TextOutput | null = null;\n  private _audioEnabled: boolean = true;\n  private _transcriptionEnabled: boolean = true;\n\n  constructor(\n    private readonly audioChanged: () => void,\n    private readonly transcriptionChanged: () => void,\n  ) {}\n\n  setAudioEnabled(enabled: boolean): void {\n    if (enabled === this._audioEnabled) {\n      return;\n    }\n\n    this._audioEnabled = enabled;\n\n    if (!this._audioSink) {\n      return;\n    }\n\n    if (enabled) {\n      this._audioSink.onAttached();\n    } else {\n      this._audioSink.onDetached();\n    }\n  }\n\n  setTranscriptionEnabled(enabled: boolean): void {\n    if (enabled === this._transcriptionEnabled) {\n      return;\n    }\n\n    this._transcriptionEnabled = enabled;\n\n    if (!this._transcriptionSink) {\n      return;\n    }\n\n    if (enabled) {\n      this._transcriptionSink.onAttached();\n    } else {\n      this._transcriptionSink.onDetached();\n    }\n  }\n\n  get audioEnabled(): boolean {\n    return this._audioEnabled;\n  }\n\n  get transcriptionEnabled(): boolean {\n    return this._transcriptionEnabled;\n  }\n\n  get audio(): AudioOutput | null {\n    return this._audioSink;\n  }\n\n  set audio(sink: AudioOutput | null) {\n    if (sink === this._audioSink) {\n      return;\n    }\n\n    if (this._audioSink) {\n      this._audioSink.onDetached();\n    }\n\n    this._audioSink = sink;\n    this.audioChanged();\n\n    if (this._audioSink) {\n      this._audioSink.onAttached();\n    }\n  }\n\n  get transcription(): TextOutput | null {\n    return this._transcriptionSink;\n  }\n\n  set transcription(sink: TextOutput | null) {\n    if (sink === this._transcriptionSink) {\n      return;\n    }\n\n    if (this._transcriptionSink) {\n      this._transcriptionSink.onDetached();\n    }\n\n    this._transcriptionSink = sink;\n    this.transcriptionChanged();\n\n    if (this._transcriptionSink) {\n      this._transcriptionSink.onAttached();\n    }\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAIA,yBAA6B;AAK7B,iBAAoB;AACpB,6BAAuC;AAEvC,mBAAuB;AAsBhB,MAAM,sBAAsB,OAAO,IAAI,gBAAgB;AAiBvD,SAAS,kBAAkB,MAMlB;AACd,SAAO;AAAA,IACL,CAAC,mBAAmB,GAAG;AAAA,IACvB,MAAM,KAAK;AAAA,IACX,WAAW,KAAK;AAAA,IAChB,SAAS,KAAK;AAAA,IACd,YAAY,KAAK;AAAA,IACjB,iBAAiB,KAAK;AAAA,EACxB;AACF;AAKO,SAAS,cAAc,OAAsC;AAClE,SACE,OAAO,UAAU,YACjB,UAAU,QACV,uBAAuB,SACtB,MAAsB,mBAAmB,MAAM;AAEpD;AAOO,MAAe,WAAW;AAAA,EACrB,iBACR,IAAI,8CAAmC;AAAA,EAEzC,IAAI,SAAqC;AACvC,WAAO,KAAK,eAAe;AAAA,EAC7B;AAAA,EAEA,aAAmB;AAAA,EAAC;AAAA,EAEpB,aAAmB;AAAA,EAAC;AACtB;AAEO,MAAe,oBAAoB,gCAAa;AAAA,EAerD,YACS,YACY,aACnB,eAAwC,EAAE,OAAO,MAAM,GACvD;AACA,UAAM;AAJC;AACY;AAInB,SAAK,eAAe;AAEpB,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY;AAAA,QAAG,YAAY;AAAA,QAAwB,CAAC,OACvD,KAAK,kBAAkB,GAAG,SAAS;AAAA,MACrC;AACA,WAAK,YAAY;AAAA,QAAG,YAAY;AAAA,QAAyB,CAAC,OACxD,KAAK,mBAAmB,EAAE;AAAA,MAC5B;AAAA,IACF;AAAA,EACF;AAAA,EA9BA,OAAgB,yBAAyB;AAAA,EACzC,OAAgB,0BAA0B;AAAA,EAElC,yBAAuC,IAAI,oBAAO;AAAA,EAClD,aAAsB;AAAA,EACtB,wBAAgC;AAAA,EAChC,wBAAgC;AAAA,EAChC,oBAA2C;AAAA,IACjD,kBAAkB;AAAA,IAClB,aAAa;AAAA,EACf;AAAA,EACU,aAAS,gBAAI;AAAA,EACJ;AAAA;AAAA;AAAA;AAAA,EAuBnB,IAAI,WAAoB;AAtI1B;AAuII,WAAO,KAAK,aAAa,YAAU,UAAK,gBAAL,mBAAkB,aAAY;AAAA,EACnE;AAAA;AAAA;AAAA;AAAA,EAKA,MAAM,aAAa,QAAmC;AACpD,QAAI,CAAC,KAAK,YAAY;AACpB,WAAK,aAAa;AAClB,WAAK;AAAA,IACP;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,MAAM,iBAAiD;AACrD,UAAM,SAAS,KAAK;AAEpB,WAAO,KAAK,wBAAwB,QAAQ;AAC1C,YAAM,KAAK,uBAAuB;AAClC,WAAK,yBAAyB,IAAI,oBAAO;AAAA,IAC3C;AAEA,WAAO,KAAK;AAAA,EACd;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,kBAAkB,WAAyB;AACzC,SAAK,KAAK,YAAY,wBAAwB,EAAE,UAAU,CAAyB;AAAA,EACrF;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,mBAAmB,SAAgC;AACjD,QAAI,KAAK,yBAAyB,KAAK,uBAAuB;AAC5D,WAAK,OAAO,KAAK,0EAA0E;AAC3F;AAAA,IACF;AAEA,SAAK,oBAAoB;AACzB,SAAK;AACL,SAAK,uBAAuB,QAAQ;AACpC,SAAK,KAAK,YAAY,yBAAyB,OAAO;AAAA,EACxD;AAAA,EAEA,QAAc;AACZ,SAAK,aAAa;AAAA,EACpB;AAAA,EAOA,aAAmB;AACjB,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY,WAAW;AAAA,IAC9B;AAAA,EACF;AAAA,EAEA,aAAmB;AACjB,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY,WAAW;AAAA,IAC9B;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,QAAc;AACZ,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY,MAAM;AAAA,IACzB;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,SAAe;AACb,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY,OAAO;AAAA,IAC1B;AAAA,EACF;AACF;AAmBO,MAAe,WAAW;AAAA,EAC/B,YAA+B,aAA0B;AAA1B;AAAA,EAA2B;AAAA,EAS1D,aAAmB;AACjB,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY,WAAW;AAAA,IAC9B;AAAA,EACF;AAAA,EAEA,aAAmB;AACjB,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY,WAAW;AAAA,IAC9B;AAAA,EACF;AACF;AAEO,MAAM,WAAW;AAAA,EAKtB,YAA6B,cAA0B;AAA1B;AAAA,EAA2B;AAAA,EAJhD,eAAkC;AAAA;AAAA,EAElC,gBAAyB;AAAA,EAIjC,gBAAgB,QAAuB;AACrC,QAAI,WAAW,KAAK,eAAe;AACjC;AAAA,IACF;AAEA,SAAK,gBAAgB;AAErB,QAAI,CAAC,KAAK,cAAc;AACtB;AAAA,IACF;AAEA,QAAI,QAAQ;AACV,WAAK,aAAa,WAAW;AAAA,IAC/B,OAAO;AACL,WAAK,aAAa,WAAW;AAAA,IAC/B;AAAA,EACF;AAAA,EAEA,IAAI,eAAwB;AAC1B,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,IAAI,QAA2B;AAC7B,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,IAAI,MAAM,QAA2B;AACnC,SAAK,eAAe;AACpB,SAAK,aAAa;AAAA,EACpB;AACF;AAEO,MAAM,YAAY;AAAA,EAMvB,YACmB,cACA,sBACjB;AAFiB;AACA;AAAA,EAChB;AAAA,EARK,aAAiC;AAAA,EACjC,qBAAwC;AAAA,EACxC,gBAAyB;AAAA,EACzB,wBAAiC;AAAA,EAOzC,gBAAgB,SAAwB;AACtC,QAAI,YAAY,KAAK,eAAe;AAClC;AAAA,IACF;AAEA,SAAK,gBAAgB;AAErB,QAAI,CAAC,KAAK,YAAY;AACpB;AAAA,IACF;AAEA,QAAI,SAAS;AACX,WAAK,WAAW,WAAW;AAAA,IAC7B,OAAO;AACL,WAAK,WAAW,WAAW;AAAA,IAC7B;AAAA,EACF;AAAA,EAEA,wBAAwB,SAAwB;AAC9C,QAAI,YAAY,KAAK,uBAAuB;AAC1C;AAAA,IACF;AAEA,SAAK,wBAAwB;AAE7B,QAAI,CAAC,KAAK,oBAAoB;AAC5B;AAAA,IACF;AAEA,QAAI,SAAS;AACX,WAAK,mBAAmB,WAAW;AAAA,IACrC,OAAO;AACL,WAAK,mBAAmB,WAAW;AAAA,IACrC;AAAA,EACF;AAAA,EAEA,IAAI,eAAwB;AAC1B,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,IAAI,uBAAgC;AAClC,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,IAAI,QAA4B;AAC9B,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,IAAI,MAAM,MAA0B;AAClC,QAAI,SAAS,KAAK,YAAY;AAC5B;AAAA,IACF;AAEA,QAAI,KAAK,YAAY;AACnB,WAAK,WAAW,WAAW;AAAA,IAC7B;AAEA,SAAK,aAAa;AAClB,SAAK,aAAa;AAElB,QAAI,KAAK,YAAY;AACnB,WAAK,WAAW,WAAW;AAAA,IAC7B;AAAA,EACF;AAAA,EAEA,IAAI,gBAAmC;AACrC,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,IAAI,cAAc,MAAyB;AACzC,QAAI,SAAS,KAAK,oBAAoB;AACpC;AAAA,IACF;AAEA,QAAI,KAAK,oBAAoB;AAC3B,WAAK,mBAAmB,WAAW;AAAA,IACrC;AAEA,SAAK,qBAAqB;AAC1B,SAAK,qBAAqB;AAE1B,QAAI,KAAK,oBAAoB;AAC3B,WAAK,mBAAmB,WAAW;AAAA,IACrC;AAAA,EACF;AACF;","names":[]}
+{"version":3,"sources":["../../src/voice/io.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2025 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport type { AudioFrame } from '@livekit/rtc-node';\nimport { EventEmitter } from 'node:events';\nimport type { ReadableStream } from 'node:stream/web';\nimport type { ChatContext } from '../llm/chat_context.js';\nimport type { ChatChunk } from '../llm/llm.js';\nimport type { ToolContext } from '../llm/tool_context.js';\nimport { log } from '../log.js';\nimport { MultiInputStream } from '../stream/multi_input_stream.js';\nimport type { SpeechEvent } from '../stt/stt.js';\nimport { Future } from '../utils.js';\nimport type { ModelSettings } from './agent.js';\n\nexport type STTNode = (\n  audio: ReadableStream<AudioFrame>,\n  modelSettings: ModelSettings,\n) => Promise<ReadableStream<SpeechEvent | string> | null>;\n\nexport type LLMNode = (\n  chatCtx: ChatContext,\n  toolCtx: ToolContext,\n  modelSettings: ModelSettings,\n) => Promise<ReadableStream<ChatChunk | string> | null>;\n\nexport type TTSNode = (\n  text: ReadableStream<string>,\n  modelSettings: ModelSettings,\n) => Promise<ReadableStream<AudioFrame> | null>;\n\n/**\n * Symbol used to identify TimedString objects.\n */\nexport const TIMED_STRING_SYMBOL = Symbol.for('lk.TimedString');\n\n/**\n * A string with optional start and end timestamps for word-level alignment.\n */\nexport interface TimedString {\n  readonly [TIMED_STRING_SYMBOL]: true;\n  text: string;\n  startTime?: number; // seconds\n  endTime?: number; // seconds\n  confidence?: number;\n  startTimeOffset?: number;\n}\n\n/**\n * Factory function to create a TimedString object.\n */\nexport function createTimedString(opts: {\n  text: string;\n  startTime?: number;\n  endTime?: number;\n  confidence?: number;\n  startTimeOffset?: number;\n}): TimedString {\n  return {\n    [TIMED_STRING_SYMBOL]: true,\n    text: opts.text,\n    startTime: opts.startTime,\n    endTime: opts.endTime,\n    confidence: opts.confidence,\n    startTimeOffset: opts.startTimeOffset,\n  };\n}\n\n/**\n * Type guard to check if a value is a TimedString.\n */\nexport function isTimedString(value: unknown): value is TimedString {\n  return (\n    typeof value === 'object' &&\n    value !== null &&\n    TIMED_STRING_SYMBOL in value &&\n    (value as TimedString)[TIMED_STRING_SYMBOL] === true\n  );\n}\n\nexport interface AudioOutputCapabilities {\n  /** Whether this output supports pause/resume functionality */\n  pause: boolean;\n}\n\nexport abstract class AudioInput {\n  protected multiStream: MultiInputStream<AudioFrame> = new MultiInputStream<AudioFrame>();\n\n  get stream(): ReadableStream<AudioFrame> {\n    return this.multiStream.stream;\n  }\n\n  async close(): Promise<void> {\n    await this.multiStream.close();\n  }\n\n  onAttached(): void {}\n\n  onDetached(): void {}\n}\n\nexport abstract class AudioOutput extends EventEmitter {\n  static readonly EVENT_PLAYBACK_STARTED = 'playbackStarted';\n  static readonly EVENT_PLAYBACK_FINISHED = 'playbackFinished';\n\n  private playbackFinishedFuture: Future<void> = new Future();\n  private _capturing: boolean = false;\n  private playbackFinishedCount: number = 0;\n  private playbackSegmentsCount: number = 0;\n  private lastPlaybackEvent: PlaybackFinishedEvent = {\n    playbackPosition: 0,\n    interrupted: false,\n  };\n  protected logger = log();\n  protected readonly capabilities: AudioOutputCapabilities;\n\n  constructor(\n    public sampleRate?: number,\n    protected readonly nextInChain?: AudioOutput,\n    capabilities: AudioOutputCapabilities = { pause: false },\n  ) {\n    super();\n    this.capabilities = capabilities;\n\n    if (this.nextInChain) {\n      this.nextInChain.on(AudioOutput.EVENT_PLAYBACK_STARTED, (ev: PlaybackStartedEvent) =>\n        this.onPlaybackStarted(ev.createdAt),\n      );\n      this.nextInChain.on(AudioOutput.EVENT_PLAYBACK_FINISHED, (ev: PlaybackFinishedEvent) =>\n        this.onPlaybackFinished(ev),\n      );\n    }\n  }\n\n  /**\n   * Whether this output and all outputs in the chain support pause/resume.\n   */\n  get canPause(): boolean {\n    return this.capabilities.pause && (this.nextInChain?.canPause ?? true);\n  }\n\n  /**\n   * Capture an audio frame for playback, frames can be pushed faster than real-time\n   */\n  async captureFrame(_frame: AudioFrame): Promise<void> {\n    if (!this._capturing) {\n      this._capturing = true;\n      this.playbackSegmentsCount++;\n    }\n  }\n\n  /**\n   * Wait for the past audio segments to finish playing out.\n   *\n   * @returns The event that was emitted when the audio finished playing out (only the last segment information)\n   */\n  async waitForPlayout(): Promise<PlaybackFinishedEvent> {\n    const target = this.playbackSegmentsCount;\n\n    while (this.playbackFinishedCount < target) {\n      await this.playbackFinishedFuture.await;\n      this.playbackFinishedFuture = new Future();\n    }\n\n    return this.lastPlaybackEvent;\n  }\n\n  /**\n   * Called when playback actually starts (first frame is sent to output).\n   * Developers building audio sinks should call this when the first frame is captured.\n   */\n  onPlaybackStarted(createdAt: number): void {\n    this.emit(AudioOutput.EVENT_PLAYBACK_STARTED, { createdAt } as PlaybackStartedEvent);\n  }\n\n  /**\n   * Developers building audio sinks must call this method when a playback/segment is finished.\n   * Segments are segmented by calls to flush() or clearBuffer()\n   */\n  onPlaybackFinished(options: PlaybackFinishedEvent) {\n    if (this.playbackFinishedCount >= this.playbackSegmentsCount) {\n      this.logger.warn('playback_finished called more times than playback segments were captured');\n      return;\n    }\n\n    this.lastPlaybackEvent = options;\n    this.playbackFinishedCount++;\n    this.playbackFinishedFuture.resolve();\n    this.emit(AudioOutput.EVENT_PLAYBACK_FINISHED, options);\n  }\n\n  flush(): void {\n    this._capturing = false;\n  }\n\n  /**\n   * Clear the buffer, stopping playback immediately\n   */\n  abstract clearBuffer(): void;\n\n  onAttached(): void {\n    if (this.nextInChain) {\n      this.nextInChain.onAttached();\n    }\n  }\n\n  onDetached(): void {\n    if (this.nextInChain) {\n      this.nextInChain.onDetached();\n    }\n  }\n\n  /**\n   * Pause the audio playback\n   */\n  pause(): void {\n    if (this.nextInChain) {\n      this.nextInChain.pause();\n    }\n  }\n\n  /**\n   * Resume the audio playback\n   */\n  resume(): void {\n    if (this.nextInChain) {\n      this.nextInChain.resume();\n    }\n  }\n}\n\nexport interface PlaybackFinishedEvent {\n  /** How much of the audio was played back, in seconds */\n  playbackPosition: number;\n  /** True if playback was interrupted (clearBuffer() was called) */\n  interrupted: boolean;\n  /**\n   * Transcript synced with playback; may be partial if the audio was interrupted.\n   * When undefined, the transcript is not synchronized with the playback.\n   */\n  synchronizedTranscript?: string;\n}\n\nexport interface PlaybackStartedEvent {\n  /** The timestamp (Date.now()) when the playback started */\n  createdAt: number;\n}\n\nexport abstract class TextOutput {\n  constructor(protected readonly nextInChain?: TextOutput) {}\n\n  abstract captureText(text: string | TimedString): Promise<void>;\n\n  /**\n   * Mark the current text segment as complete (e.g LLM generation is complete)\n   */\n  abstract flush(): void;\n\n  onAttached(): void {\n    if (this.nextInChain) {\n      this.nextInChain.onAttached();\n    }\n  }\n\n  onDetached(): void {\n    if (this.nextInChain) {\n      this.nextInChain.onDetached();\n    }\n  }\n}\n\nexport class AgentInput {\n  private _audioStream: AudioInput | null = null;\n  // enabled by default\n  private _audioEnabled: boolean = true;\n\n  constructor(private readonly audioChanged: () => void) {}\n\n  setAudioEnabled(enable: boolean): void {\n    if (enable === this._audioEnabled) {\n      return;\n    }\n\n    this._audioEnabled = enable;\n\n    if (!this._audioStream) {\n      return;\n    }\n\n    if (enable) {\n      this._audioStream.onAttached();\n    } else {\n      this._audioStream.onDetached();\n    }\n  }\n\n  get audioEnabled(): boolean {\n    return this._audioEnabled;\n  }\n\n  get audio(): AudioInput | null {\n    return this._audioStream;\n  }\n\n  set audio(stream: AudioInput | null) {\n    this._audioStream = stream;\n    this.audioChanged();\n  }\n}\n\nexport class AgentOutput {\n  private _audioSink: AudioOutput | null = null;\n  private _transcriptionSink: TextOutput | null = null;\n  private _audioEnabled: boolean = true;\n  private _transcriptionEnabled: boolean = true;\n\n  constructor(\n    private readonly audioChanged: () => void,\n    private readonly transcriptionChanged: () => void,\n  ) {}\n\n  setAudioEnabled(enabled: boolean): void {\n    if (enabled === this._audioEnabled) {\n      return;\n    }\n\n    this._audioEnabled = enabled;\n\n    if (!this._audioSink) {\n      return;\n    }\n\n    if (enabled) {\n      this._audioSink.onAttached();\n    } else {\n      this._audioSink.onDetached();\n    }\n  }\n\n  setTranscriptionEnabled(enabled: boolean): void {\n    if (enabled === this._transcriptionEnabled) {\n      return;\n    }\n\n    this._transcriptionEnabled = enabled;\n\n    if (!this._transcriptionSink) {\n      return;\n    }\n\n    if (enabled) {\n      this._transcriptionSink.onAttached();\n    } else {\n      this._transcriptionSink.onDetached();\n    }\n  }\n\n  get audioEnabled(): boolean {\n    return this._audioEnabled;\n  }\n\n  get transcriptionEnabled(): boolean {\n    return this._transcriptionEnabled;\n  }\n\n  get audio(): AudioOutput | null {\n    return this._audioSink;\n  }\n\n  set audio(sink: AudioOutput | null) {\n    if (sink === this._audioSink) {\n      return;\n    }\n\n    if (this._audioSink) {\n      this._audioSink.onDetached();\n    }\n\n    this._audioSink = sink;\n    this.audioChanged();\n\n    if (this._audioSink) {\n      this._audioSink.onAttached();\n    }\n  }\n\n  get transcription(): TextOutput | null {\n    return this._transcriptionSink;\n  }\n\n  set transcription(sink: TextOutput | null) {\n    if (sink === this._transcriptionSink) {\n      return;\n    }\n\n    if (this._transcriptionSink) {\n      this._transcriptionSink.onDetached();\n    }\n\n    this._transcriptionSink = sink;\n    this.transcriptionChanged();\n\n    if (this._transcriptionSink) {\n      this._transcriptionSink.onAttached();\n    }\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAIA,yBAA6B;AAK7B,iBAAoB;AACpB,gCAAiC;AAEjC,mBAAuB;AAsBhB,MAAM,sBAAsB,OAAO,IAAI,gBAAgB;AAiBvD,SAAS,kBAAkB,MAMlB;AACd,SAAO;AAAA,IACL,CAAC,mBAAmB,GAAG;AAAA,IACvB,MAAM,KAAK;AAAA,IACX,WAAW,KAAK;AAAA,IAChB,SAAS,KAAK;AAAA,IACd,YAAY,KAAK;AAAA,IACjB,iBAAiB,KAAK;AAAA,EACxB;AACF;AAKO,SAAS,cAAc,OAAsC;AAClE,SACE,OAAO,UAAU,YACjB,UAAU,QACV,uBAAuB,SACtB,MAAsB,mBAAmB,MAAM;AAEpD;AAOO,MAAe,WAAW;AAAA,EACrB,cAA4C,IAAI,2CAA6B;AAAA,EAEvF,IAAI,SAAqC;AACvC,WAAO,KAAK,YAAY;AAAA,EAC1B;AAAA,EAEA,MAAM,QAAuB;AAC3B,UAAM,KAAK,YAAY,MAAM;AAAA,EAC/B;AAAA,EAEA,aAAmB;AAAA,EAAC;AAAA,EAEpB,aAAmB;AAAA,EAAC;AACtB;AAEO,MAAe,oBAAoB,gCAAa;AAAA,EAerD,YACS,YACY,aACnB,eAAwC,EAAE,OAAO,MAAM,GACvD;AACA,UAAM;AAJC;AACY;AAInB,SAAK,eAAe;AAEpB,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY;AAAA,QAAG,YAAY;AAAA,QAAwB,CAAC,OACvD,KAAK,kBAAkB,GAAG,SAAS;AAAA,MACrC;AACA,WAAK,YAAY;AAAA,QAAG,YAAY;AAAA,QAAyB,CAAC,OACxD,KAAK,mBAAmB,EAAE;AAAA,MAC5B;AAAA,IACF;AAAA,EACF;AAAA,EA9BA,OAAgB,yBAAyB;AAAA,EACzC,OAAgB,0BAA0B;AAAA,EAElC,yBAAuC,IAAI,oBAAO;AAAA,EAClD,aAAsB;AAAA,EACtB,wBAAgC;AAAA,EAChC,wBAAgC;AAAA,EAChC,oBAA2C;AAAA,IACjD,kBAAkB;AAAA,IAClB,aAAa;AAAA,EACf;AAAA,EACU,aAAS,gBAAI;AAAA,EACJ;AAAA;AAAA;AAAA;AAAA,EAuBnB,IAAI,WAAoB;AAzI1B;AA0II,WAAO,KAAK,aAAa,YAAU,UAAK,gBAAL,mBAAkB,aAAY;AAAA,EACnE;AAAA;AAAA;AAAA;AAAA,EAKA,MAAM,aAAa,QAAmC;AACpD,QAAI,CAAC,KAAK,YAAY;AACpB,WAAK,aAAa;AAClB,WAAK;AAAA,IACP;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAOA,MAAM,iBAAiD;AACrD,UAAM,SAAS,KAAK;AAEpB,WAAO,KAAK,wBAAwB,QAAQ;AAC1C,YAAM,KAAK,uBAAuB;AAClC,WAAK,yBAAyB,IAAI,oBAAO;AAAA,IAC3C;AAEA,WAAO,KAAK;AAAA,EACd;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,kBAAkB,WAAyB;AACzC,SAAK,KAAK,YAAY,wBAAwB,EAAE,UAAU,CAAyB;AAAA,EACrF;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,mBAAmB,SAAgC;AACjD,QAAI,KAAK,yBAAyB,KAAK,uBAAuB;AAC5D,WAAK,OAAO,KAAK,0EAA0E;AAC3F;AAAA,IACF;AAEA,SAAK,oBAAoB;AACzB,SAAK;AACL,SAAK,uBAAuB,QAAQ;AACpC,SAAK,KAAK,YAAY,yBAAyB,OAAO;AAAA,EACxD;AAAA,EAEA,QAAc;AACZ,SAAK,aAAa;AAAA,EACpB;AAAA,EAOA,aAAmB;AACjB,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY,WAAW;AAAA,IAC9B;AAAA,EACF;AAAA,EAEA,aAAmB;AACjB,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY,WAAW;AAAA,IAC9B;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,QAAc;AACZ,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY,MAAM;AAAA,IACzB;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,SAAe;AACb,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY,OAAO;AAAA,IAC1B;AAAA,EACF;AACF;AAmBO,MAAe,WAAW;AAAA,EAC/B,YAA+B,aAA0B;AAA1B;AAAA,EAA2B;AAAA,EAS1D,aAAmB;AACjB,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY,WAAW;AAAA,IAC9B;AAAA,EACF;AAAA,EAEA,aAAmB;AACjB,QAAI,KAAK,aAAa;AACpB,WAAK,YAAY,WAAW;AAAA,IAC9B;AAAA,EACF;AACF;AAEO,MAAM,WAAW;AAAA,EAKtB,YAA6B,cAA0B;AAA1B;AAAA,EAA2B;AAAA,EAJhD,eAAkC;AAAA;AAAA,EAElC,gBAAyB;AAAA,EAIjC,gBAAgB,QAAuB;AACrC,QAAI,WAAW,KAAK,eAAe;AACjC;AAAA,IACF;AAEA,SAAK,gBAAgB;AAErB,QAAI,CAAC,KAAK,cAAc;AACtB;AAAA,IACF;AAEA,QAAI,QAAQ;AACV,WAAK,aAAa,WAAW;AAAA,IAC/B,OAAO;AACL,WAAK,aAAa,WAAW;AAAA,IAC/B;AAAA,EACF;AAAA,EAEA,IAAI,eAAwB;AAC1B,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,IAAI,QAA2B;AAC7B,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,IAAI,MAAM,QAA2B;AACnC,SAAK,eAAe;AACpB,SAAK,aAAa;AAAA,EACpB;AACF;AAEO,MAAM,YAAY;AAAA,EAMvB,YACmB,cACA,sBACjB;AAFiB;AACA;AAAA,EAChB;AAAA,EARK,aAAiC;AAAA,EACjC,qBAAwC;AAAA,EACxC,gBAAyB;AAAA,EACzB,wBAAiC;AAAA,EAOzC,gBAAgB,SAAwB;AACtC,QAAI,YAAY,KAAK,eAAe;AAClC;AAAA,IACF;AAEA,SAAK,gBAAgB;AAErB,QAAI,CAAC,KAAK,YAAY;AACpB;AAAA,IACF;AAEA,QAAI,SAAS;AACX,WAAK,WAAW,WAAW;AAAA,IAC7B,OAAO;AACL,WAAK,WAAW,WAAW;AAAA,IAC7B;AAAA,EACF;AAAA,EAEA,wBAAwB,SAAwB;AAC9C,QAAI,YAAY,KAAK,uBAAuB;AAC1C;AAAA,IACF;AAEA,SAAK,wBAAwB;AAE7B,QAAI,CAAC,KAAK,oBAAoB;AAC5B;AAAA,IACF;AAEA,QAAI,SAAS;AACX,WAAK,mBAAmB,WAAW;AAAA,IACrC,OAAO;AACL,WAAK,mBAAmB,WAAW;AAAA,IACrC;AAAA,EACF;AAAA,EAEA,IAAI,eAAwB;AAC1B,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,IAAI,uBAAgC;AAClC,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,IAAI,QAA4B;AAC9B,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,IAAI,MAAM,MAA0B;AAClC,QAAI,SAAS,KAAK,YAAY;AAC5B;AAAA,IACF;AAEA,QAAI,KAAK,YAAY;AACnB,WAAK,WAAW,WAAW;AAAA,IAC7B;AAEA,SAAK,aAAa;AAClB,SAAK,aAAa;AAElB,QAAI,KAAK,YAAY;AACnB,WAAK,WAAW,WAAW;AAAA,IAC7B;AAAA,EACF;AAAA,EAEA,IAAI,gBAAmC;AACrC,WAAO,KAAK;AAAA,EACd;AAAA,EAEA,IAAI,cAAc,MAAyB;AACzC,QAAI,SAAS,KAAK,oBAAoB;AACpC;AAAA,IACF;AAEA,QAAI,KAAK,oBAAoB;AAC3B,WAAK,mBAAmB,WAAW;AAAA,IACrC;AAEA,SAAK,qBAAqB;AAC1B,SAAK,qBAAqB;AAE1B,QAAI,KAAK,oBAAoB;AAC3B,WAAK,mBAAmB,WAAW;AAAA,IACrC;AAAA,EACF;AACF;","names":[]}

package/dist/voice/io.d.cts

@@ -6,7 +6,7 @@ import type { ReadableStream } from 'node:stream/web';
 import type { ChatContext } from '../llm/chat_context.js';
 import type { ChatChunk } from '../llm/llm.js';
 import type { ToolContext } from '../llm/tool_context.js';
-import { DeferredReadableStream } from '../stream/deferred_stream.js';
+import { MultiInputStream } from '../stream/multi_input_stream.js';
 import type { SpeechEvent } from '../stt/stt.js';
 import type { ModelSettings } from './agent.js';
 export type STTNode = (audio: ReadableStream<AudioFrame>, modelSettings: ModelSettings) => Promise<ReadableStream<SpeechEvent | string> | null>;
@@ -46,8 +46,9 @@ export interface AudioOutputCapabilities {
     pause: boolean;
 }
 export declare abstract class AudioInput {
-    protected deferredStream: DeferredReadableStream<AudioFrame>;
+    protected multiStream: MultiInputStream<AudioFrame>;
     get stream(): ReadableStream<AudioFrame>;
+    close(): Promise<void>;
     onAttached(): void;
     onDetached(): void;
 }

package/dist/voice/io.d.ts

@@ -6,7 +6,7 @@ import type { ReadableStream } from 'node:stream/web';
 import type { ChatContext } from '../llm/chat_context.js';
 import type { ChatChunk } from '../llm/llm.js';
 import type { ToolContext } from '../llm/tool_context.js';
-import { DeferredReadableStream } from '../stream/deferred_stream.js';
+import { MultiInputStream } from '../stream/multi_input_stream.js';
 import type { SpeechEvent } from '../stt/stt.js';
 import type { ModelSettings } from './agent.js';
 export type STTNode = (audio: ReadableStream<AudioFrame>, modelSettings: ModelSettings) => Promise<ReadableStream<SpeechEvent | string> | null>;
@@ -46,8 +46,9 @@ export interface AudioOutputCapabilities {
     pause: boolean;
 }
 export declare abstract class AudioInput {
-    protected deferredStream: DeferredReadableStream<AudioFrame>;
+    protected multiStream: MultiInputStream<AudioFrame>;
     get stream(): ReadableStream<AudioFrame>;
+    close(): Promise<void>;
     onAttached(): void;
     onDetached(): void;
 }