fix: cancel the pending speech-end timer when a call is torn down

__tests__/speaking-timeout.test.ts

@@ -0,0 +1,61 @@
+import Vapi from "../vapi";
+
+jest.mock("@daily-co/daily-js", () => ({
+  __esModule: true,
+  default: { createCallObject: () => null },
+}));
+
+jest.mock("../client", () => ({
+  client: { baseUrl: "", setSecurityData: () => {}, call: {} },
+}));
+
+const speaking = { participantsAudioLevel: { assistant: 0.1 } } as any;
+
+describe("speech-end timer", () => {
+  let vapi: Vapi;
+  let events: string[];
+
+  beforeEach(() => {
+    jest.useFakeTimers();
+    vapi = new Vapi("dummy_token");
+    vapi["call"] = { destroy: jest.fn().mockResolvedValue(undefined) } as any;
+    events = [];
+    vapi.on("speech-start", () => events.push("speech-start"));
+    vapi.on("speech-end", () => events.push("speech-end"));
+  });
+
+  afterEach(() => {
+    jest.useRealTimers();
+  });
+
+  it("emits speech-end once the assistant stops speaking", () => {
+    vapi["handleRemoteParticipantsAudioLevel"](speaking);
+    jest.advanceTimersByTime(1000);
+    expect(events).toEqual(["speech-start", "speech-end"]);
+  });
+
+  it("does not fire a stale speech-end into the next call after stop()", async () => {
+    vapi["handleRemoteParticipantsAudioLevel"](speaking);
+    await vapi.stop();
+    events.length = 0;
+
+    // Next call: the assistant starts speaking 500 ms later.
+    jest.advanceTimersByTime(500);
+    vapi["handleRemoteParticipantsAudioLevel"](speaking);
+    jest.advanceTimersByTime(600);
+
+    // The old timer must not end the new call's speech early.
+    expect(events).toEqual(["speech-start"]);
+    jest.advanceTimersByTime(500);
+    expect(events).toEqual(["speech-start", "speech-end"]);
+  });
+
+  it("emits speech-end when the call is torn down while the assistant speaks", async () => {
+    vapi["handleRemoteParticipantsAudioLevel"](speaking);
+    events.length = 0;
+    await vapi.stop();
+    expect(events).toEqual(["speech-end"]);
+    jest.advanceTimersByTime(5000);
+    expect(events).toEqual(["speech-end"]);
+  });
+});

vapi.ts

@@ -384,7 +384,7 @@ export default class Vapi extends VapiEventEmitter {
     }
     this.audioPlayer = null;
     this.pendingAudioBuilds.clear();
-    this.speakingTimeout = null;
+    this.endSpeaking();
   }
 
   private isMobileDevice() {
@@ -1081,6 +1081,19 @@ export default class Vapi extends VapiEventEmitter {
     }, 1000);
   }
 
+  /**
+   * Cancels the pending speech-end timer. If the assistant was mid-speech,
+   * emits speech-end now, so it is not fired later into the next call.
+   */
+  private endSpeaking() {
+    if (!this.speakingTimeout) {
+      return;
+    }
+    clearTimeout(this.speakingTimeout);
+    this.speakingTimeout = null;
+    this.emit('speech-end');
+  }
+
   private handleLocalAudioLevel(e: DailyEventObjectLocalAudioLevel) {
     this.emit('local-volume-level', e.audioLevel);
   }
@@ -1099,7 +1112,7 @@ export default class Vapi extends VapiEventEmitter {
     }
     this.audioPlayer = null;
     this.pendingAudioBuilds.clear();
-    this.speakingTimeout = null;
+    this.endSpeaking();
   }
 
   /**