diff --git a/ui/desktop/src/hooks/useAudioRecorder.test.tsx b/ui/desktop/src/hooks/useAudioRecorder.test.tsx
new file mode 100644
index 000000000..8c3fe1768
--- /dev/null
+++ b/ui/desktop/src/hooks/useAudioRecorder.test.tsx
@@ -0,0 +1,348 @@
+import { act, renderHook, waitFor } from '@testing-library/react';
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
+
+const mocks = vi.hoisted(() => ({
+  config: {},
+  getDictationConfig: vi.fn(),
+  onError: vi.fn(),
+  onTranscription: vi.fn(),
+  read: vi.fn(),
+  transcribeDictation: vi.fn(),
+}));
+
+vi.mock('../components/ConfigContext', () => ({
+  useConfig: () => ({ config: mocks.config, read: mocks.read }),
+}));
+
+vi.mock('../acp/dictation', () => ({
+  getDictationConfig: mocks.getDictationConfig,
+  transcribeDictation: mocks.transcribeDictation,
+}));
+
+import { useAudioRecorder } from './useAudioRecorder';
+
+type Deferred<T> = {
+  promise: Promise<T>;
+  resolve: (value: T) => void;
+};
+
+type FakeStream = MediaStream & {
+  track: { stop: ReturnType<typeof vi.fn> };
+};
+
+const deferred = <T,>(): Deferred<T> => {
+  let resolve = (_value: T) => {};
+  const promise = new Promise<T>((resolvePromise) => {
+    resolve = resolvePromise;
+  });
+  return { promise, resolve };
+};
+
+const createStream = (): FakeStream => {
+  const track = { stop: vi.fn() };
+  return {
+    track,
+    getTracks: () => [track],
+  } as unknown as FakeStream;
+};
+
+let moduleLoads: Deferred<void>[];
+let audioContexts: MockAudioContext[];
+let worklets: MockAudioWorkletNode[];
+
+class MockAudioContext {
+  audioWorklet: { addModule: ReturnType<typeof vi.fn> };
+  close = vi.fn(() => Promise.resolve());
+  createGain = vi.fn(() => ({
+    connect: vi.fn(),
+    gain: { value: 1 },
+  }));
+  createMediaStreamSource = vi.fn(() => ({ connect: vi.fn() }));
+  destination = {};
+
+  constructor() {
+    const moduleLoad = moduleLoads.shift();
+    this.audioWorklet = {
+      addModule: vi.fn(() => moduleLoad?.promise ?? Promise.resolve()),
+    };
+    audioContexts.push(this);
+  }
+}
+
+class MockAudioWorkletNode {
+  connect = vi.fn();
+  disconnect = vi.fn();
+  port: { onmessage: ((event: MessageEvent<Float32Array>) => void) | null } = {
+    onmessage: null,
+  };
+
+  constructor() {
+    worklets.push(this);
+  }
+
+  emit(samples: Float32Array) {
+    this.port.onmessage?.({ data: samples } as MessageEvent<Float32Array>);
+  }
+}
+
+const renderRecorder = async () => {
+  const hook = renderHook(() =>
+    useAudioRecorder({
+      onError: mocks.onError,
+      onTranscription: mocks.onTranscription,
+    })
+  );
+  await waitFor(() => expect(hook.result.current.isEnabled).toBe(true));
+  return hook;
+};
+
+const startRecorder = async (startRecording: () => Promise<void>) => {
+  await act(async () => {
+    await startRecording();
+  });
+};
+
+const emitSpeechChunk = (worklet: MockAudioWorkletNode) => {
+  let now = 100;
+  vi.spyOn(Date, 'now').mockImplementation(() => now);
+  const speech = new Float32Array(3200).fill(0.1);
+  const silence = new Float32Array(3200);
+
+  act(() => {
+    worklet.emit(speech);
+    now = 400;
+    worklet.emit(silence);
+    now = 1301;
+    worklet.emit(silence);
+  });
+};
+
+describe('useAudioRecorder lifecycle', () => {
+  let getUserMedia: ReturnType<typeof vi.fn>;
+
+  beforeEach(() => {
+    moduleLoads = [];
+    audioContexts = [];
+    worklets = [];
+    getUserMedia = vi.fn();
+
+    mocks.getDictationConfig.mockReset().mockResolvedValue({
+      openai: { configured: true },
+    });
+    mocks.onError.mockReset();
+    mocks.onTranscription.mockReset();
+    mocks.read
+      .mockReset()
+      .mockImplementation((key: string) =>
+        Promise.resolve(key === 'voice_dictation_provider' ? 'openai' : null)
+      );
+    mocks.transcribeDictation.mockReset();
+
+    vi.stubGlobal('AudioContext', MockAudioContext);
+    vi.stubGlobal('AudioWorkletNode', MockAudioWorkletNode);
+    Object.defineProperty(navigator, 'mediaDevices', {
+      configurable: true,
+      value: { getUserMedia },
+    });
+  });
+
+  afterEach(() => {
+    vi.restoreAllMocks();
+    vi.unstubAllGlobals();
+  });
+
+  it('cleans the superseded generation when starts overlap', async () => {
+    const firstStream = createStream();
+    const secondStream = createStream();
+    const firstModuleLoad = deferred<void>();
+    moduleLoads.push(firstModuleLoad);
+    getUserMedia.mockResolvedValueOnce(firstStream).mockResolvedValueOnce(secondStream);
+    const { result } = await renderRecorder();
+
+    let firstStart = Promise.resolve();
+    act(() => {
+      firstStart = result.current.startRecording();
+    });
+    await waitFor(() => expect(audioContexts).toHaveLength(1));
+
+    await startRecorder(result.current.startRecording);
+
+    expect(firstStream.track.stop).toHaveBeenCalledOnce();
+    expect(audioContexts[0].close).toHaveBeenCalledOnce();
+    expect(worklets).toHaveLength(1);
+    expect(result.current.isRecording).toBe(true);
+
+    await act(async () => {
+      firstModuleLoad.resolve();
+      await firstStart;
+    });
+
+    expect(worklets).toHaveLength(1);
+    act(() => result.current.stopRecording());
+    expect(secondStream.track.stop).toHaveBeenCalledOnce();
+    expect(audioContexts[1].close).toHaveBeenCalledOnce();
+    expect(worklets[0].disconnect).toHaveBeenCalledOnce();
+  });
+
+  it('stops a stream that arrives after recording is stopped', async () => {
+    const pendingStream = deferred<MediaStream>();
+    const stream = createStream();
+    getUserMedia.mockReturnValueOnce(pendingStream.promise);
+    const { result } = await renderRecorder();
+
+    let start = Promise.resolve();
+    act(() => {
+      start = result.current.startRecording();
+    });
+    await waitFor(() => expect(getUserMedia).toHaveBeenCalledOnce());
+    act(() => result.current.stopRecording());
+
+    await act(async () => {
+      pendingStream.resolve(stream);
+      await start;
+    });
+
+    expect(stream.track.stop).toHaveBeenCalledOnce();
+    expect(audioContexts).toHaveLength(0);
+    expect(worklets).toHaveLength(0);
+    expect(result.current.isRecording).toBe(false);
+    expect(mocks.onError).not.toHaveBeenCalled();
+  });
+
+  it('does not finish startup after unmount', async () => {
+    const stream = createStream();
+    const moduleLoad = deferred<void>();
+    moduleLoads.push(moduleLoad);
+    getUserMedia.mockResolvedValueOnce(stream);
+    const { result, unmount } = await renderRecorder();
+
+    let start = Promise.resolve();
+    act(() => {
+      start = result.current.startRecording();
+    });
+    await waitFor(() => expect(audioContexts).toHaveLength(1));
+    unmount();
+
+    expect(stream.track.stop).toHaveBeenCalledOnce();
+    expect(audioContexts[0].close).toHaveBeenCalledOnce();
+
+    await act(async () => {
+      moduleLoad.resolve();
+      await start;
+    });
+
+    expect(worklets).toHaveLength(0);
+    expect(mocks.onError).not.toHaveBeenCalled();
+  });
+
+  it('suppresses an in-flight transcription after stop', async () => {
+    const stream = createStream();
+    const transcription = deferred<string>();
+    getUserMedia.mockResolvedValueOnce(stream);
+    mocks.transcribeDictation.mockReturnValueOnce(transcription.promise);
+    const { result } = await renderRecorder();
+    await startRecorder(result.current.startRecording);
+
+    emitSpeechChunk(worklets[0]);
+    await waitFor(() => expect(mocks.transcribeDictation).toHaveBeenCalledOnce());
+    act(() => result.current.stopRecording());
+
+    await act(async () => {
+      transcription.resolve('stale transcription');
+      await transcription.promise;
+      await Promise.resolve();
+    });
+
+    expect(mocks.onTranscription).not.toHaveBeenCalled();
+  });
+
+  it('transcribes the buffered final phrase when recording is stopped', async () => {
+    const stream = createStream();
+    getUserMedia.mockResolvedValueOnce(stream);
+    mocks.transcribeDictation.mockResolvedValueOnce('final phrase');
+    const { result } = await renderRecorder();
+    await startRecorder(result.current.startRecording);
+
+    act(() => {
+      worklets[0].emit(new Float32Array(3200).fill(0.1));
+      result.current.stopRecording();
+    });
+
+    await waitFor(() => expect(mocks.onTranscription).toHaveBeenCalledWith('final phrase'));
+    await waitFor(() => expect(result.current.isTranscribing).toBe(false));
+    expect(stream.track.stop).toHaveBeenCalledOnce();
+    expect(audioContexts[0].close).toHaveBeenCalledOnce();
+    expect(worklets[0].disconnect).toHaveBeenCalledOnce();
+  });
+
+  it('cancels final-phrase transcription when a new recording starts', async () => {
+    const firstStream = createStream();
+    const secondStream = createStream();
+    const transcription = deferred<string>();
+    getUserMedia.mockResolvedValueOnce(firstStream).mockResolvedValueOnce(secondStream);
+    mocks.transcribeDictation.mockReturnValueOnce(transcription.promise);
+    const { result } = await renderRecorder();
+    await startRecorder(result.current.startRecording);
+
+    act(() => {
+      worklets[0].emit(new Float32Array(3200).fill(0.1));
+      result.current.stopRecording();
+    });
+    await waitFor(() => expect(mocks.transcribeDictation).toHaveBeenCalledOnce());
+
+    await startRecorder(result.current.startRecording);
+    await act(async () => {
+      transcription.resolve('stale final phrase');
+      await transcription.promise;
+      await Promise.resolve();
+    });
+
+    expect(mocks.onTranscription).not.toHaveBeenCalled();
+    expect(result.current.isRecording).toBe(true);
+    act(() => result.current.stopRecording());
+  });
+
+  it('cancels final-phrase transcription when unmounted', async () => {
+    const stream = createStream();
+    const transcription = deferred<string>();
+    getUserMedia.mockResolvedValueOnce(stream);
+    mocks.transcribeDictation.mockReturnValueOnce(transcription.promise);
+    const { result, unmount } = await renderRecorder();
+    await startRecorder(result.current.startRecording);
+
+    act(() => {
+      worklets[0].emit(new Float32Array(3200).fill(0.1));
+      result.current.stopRecording();
+    });
+    await waitFor(() => expect(mocks.transcribeDictation).toHaveBeenCalledOnce());
+    unmount();
+
+    await act(async () => {
+      transcription.resolve('stale final phrase');
+      await transcription.promise;
+      await Promise.resolve();
+    });
+
+    expect(mocks.onTranscription).not.toHaveBeenCalled();
+  });
+
+  it('records and transcribes a normal generation', async () => {
+    const stream = createStream();
+    getUserMedia.mockResolvedValueOnce(stream);
+    mocks.transcribeDictation.mockResolvedValueOnce('hello world');
+    const { result } = await renderRecorder();
+
+    await startRecorder(result.current.startRecording);
+    expect(result.current.isRecording).toBe(true);
+
+    emitSpeechChunk(worklets[0]);
+    await waitFor(() => expect(mocks.onTranscription).toHaveBeenCalledWith('hello world'));
+    await waitFor(() => expect(result.current.isTranscribing).toBe(false));
+
+    act(() => result.current.stopRecording());
+    expect(result.current.isRecording).toBe(false);
+    expect(stream.track.stop).toHaveBeenCalledOnce();
+    expect(audioContexts[0].close).toHaveBeenCalledOnce();
+    expect(worklets[0].disconnect).toHaveBeenCalledOnce();
+  });
+});
diff --git a/ui/desktop/src/hooks/useAudioRecorder.ts b/ui/desktop/src/hooks/useAudioRecorder.ts
index 4e4fe68c2..27760c3cb 100644
--- a/ui/desktop/src/hooks/useAudioRecorder.ts
+++ b/ui/desktop/src/hooks/useAudioRecorder.ts
@@ -17,6 +17,36 @@ const MIN_SPEECH_MS = 200;
 // without clipping early speech onsets. Determined empirically for 16kHz mono input.
 const RMS_THRESHOLD = 0.015;
 
+interface RecorderGeneration {
+  audioContext: AudioContext | null;
+  cancelled: boolean;
+  completesAfterTranscription: boolean;
+  pendingTranscriptions: number;
+  stream: MediaStream | null;
+  worklet: AudioWorkletNode | null;
+}
+
+function cleanupGeneration(generation: RecorderGeneration) {
+  generation.cancelled = true;
+
+  const worklet = generation.worklet;
+  generation.worklet = null;
+  if (worklet) {
+    worklet.port.onmessage = null;
+    worklet.disconnect();
+  }
+
+  const audioContext = generation.audioContext;
+  generation.audioContext = null;
+  if (audioContext) {
+    void audioContext.close();
+  }
+
+  const stream = generation.stream;
+  generation.stream = null;
+  stream?.getTracks().forEach((track) => track.stop());
+}
+
 // Resolve worklet URL at runtime from window.location so it works under both
 // the dev server (http://localhost) and packaged builds (file://).
 const WORKLET_URL = new URL('audio-capture-worklet.js', window.location.href.split('#')[0]).href;
@@ -55,6 +85,17 @@ function rms(samples: Float32Array): number {
   return Math.sqrt(sum / samples.length);
 }
 
+function mergeSamples(chunks: Float32Array[]): Float32Array {
+  const total = chunks.reduce((length, chunk) => length + chunk.length, 0);
+  const merged = new Float32Array(total);
+  let offset = 0;
+  for (const chunk of chunks) {
+    merged.set(chunk, offset);
+    offset += chunk.length;
+  }
+  return merged;
+}
+
 function blobToBase64(blob: Blob): Promise<string> {
   return new Promise((resolve, reject) => {
     const r = new FileReader();
@@ -72,15 +113,14 @@ export const useAudioRecorder = ({ onTranscription, onError }: UseAudioRecorderO
 
   const { read, config } = useConfig();
 
-  const audioContextRef = useRef<AudioContext | null>(null);
-  const streamRef = useRef<MediaStream | null>(null);
+  const activeGenerationRef = useRef<RecorderGeneration | null>(null);
+  const mountedRef = useRef(true);
 
   // VAD state (all refs to avoid re-render/stale closure issues)
   const samplesRef = useRef<Float32Array[]>([]);
   const isSpeakingRef = useRef(false);
   const silenceStartRef = useRef(0);
   const speechStartRef = useRef(0);
-  const pendingTranscriptions = useRef(0);
   const providerRef = useRef(provider);
   providerRef.current = provider;
 
@@ -112,95 +152,149 @@ export const useAudioRecorder = ({ onTranscription, onError }: UseAudioRecorderO
     check();
   }, [read, config]);
 
-  const transcribeChunk = useCallback(async (samples: Float32Array) => {
-    const prov = providerRef.current;
-    if (!prov) return;
-
-    pendingTranscriptions.current++;
-    setIsTranscribing(true);
-
-    try {
-      const wav = new Blob([encodeWav(samples, SAMPLE_RATE)], { type: 'audio/wav' });
-      const base64 = await blobToBase64(wav);
-      const text = await transcribeDictation(base64, 'audio/wav', prov);
-      if (text) {
-        onTranscriptionRef.current(text);
-      }
-    } catch (error) {
-      onErrorRef.current(errorMessage(error));
-    } finally {
-      pendingTranscriptions.current--;
-      if (pendingTranscriptions.current === 0) setIsTranscribing(false);
-    }
-  }, []);
-
-  const flush = useCallback(() => {
-    const chunks = samplesRef.current;
-    if (chunks.length === 0) return;
-
-    const total = chunks.reduce((n, c) => n + c.length, 0);
-    const merged = new Float32Array(total);
-    let off = 0;
-    for (const c of chunks) {
-      merged.set(c, off);
-      off += c.length;
-    }
+  const resetSpeech = useCallback(() => {
     samplesRef.current = [];
-    transcribeChunk(merged);
-  }, [transcribeChunk]);
-
-  const flushRef = useRef(flush);
-  flushRef.current = flush;
-
-  const handleSamples = useCallback((samples: Float32Array) => {
-    const now = Date.now();
-
-    if (rms(samples) > RMS_THRESHOLD) {
-      if (!isSpeakingRef.current) {
-        isSpeakingRef.current = true;
-        speechStartRef.current = now;
-      }
-      silenceStartRef.current = 0;
-      samplesRef.current.push(new Float32Array(samples));
-    } else if (isSpeakingRef.current) {
-      samplesRef.current.push(new Float32Array(samples));
-
-      if (silenceStartRef.current === 0) {
-        silenceStartRef.current = now;
-      } else if (now - silenceStartRef.current > SILENCE_MS) {
-        if (now - speechStartRef.current > MIN_SPEECH_MS) {
-          flushRef.current();
-        } else {
-          samplesRef.current = [];
-        }
-        isSpeakingRef.current = false;
-        silenceStartRef.current = 0;
-      }
-    }
-  }, []);
-
-  const stopRecording = useCallback(() => {
-    if (isSpeakingRef.current && samplesRef.current.length > 0) {
-      flushRef.current();
-    }
     isSpeakingRef.current = false;
     silenceStartRef.current = 0;
-
-    audioContextRef.current?.close();
-    audioContextRef.current = null;
-    streamRef.current?.getTracks().forEach((t) => t.stop());
-    streamRef.current = null;
-    setIsRecording(false);
+    speechStartRef.current = 0;
   }, []);
 
+  const isActiveGeneration = useCallback(
+    (generation: RecorderGeneration) =>
+      mountedRef.current && activeGenerationRef.current === generation && !generation.cancelled,
+    []
+  );
+
+  const transcribeChunk = useCallback(
+    async (samples: Float32Array, generation: RecorderGeneration) => {
+      const prov = providerRef.current;
+      if (!prov || !isActiveGeneration(generation)) return;
+
+      generation.pendingTranscriptions++;
+      setIsTranscribing(true);
+
+      try {
+        const wav = new Blob([encodeWav(samples, SAMPLE_RATE)], { type: 'audio/wav' });
+        const base64 = await blobToBase64(wav);
+        if (!isActiveGeneration(generation)) return;
+
+        const text = await transcribeDictation(base64, 'audio/wav', prov);
+        if (text && isActiveGeneration(generation)) {
+          onTranscriptionRef.current(text);
+        }
+      } catch (error) {
+        if (isActiveGeneration(generation)) {
+          onErrorRef.current(errorMessage(error));
+        }
+      } finally {
+        generation.pendingTranscriptions--;
+        if (generation.pendingTranscriptions === 0 && isActiveGeneration(generation)) {
+          setIsTranscribing(false);
+          if (generation.completesAfterTranscription) {
+            activeGenerationRef.current = null;
+            generation.cancelled = true;
+          }
+        }
+      }
+    },
+    [isActiveGeneration]
+  );
+
+  const flush = useCallback(
+    (generation: RecorderGeneration) => {
+      if (!isActiveGeneration(generation)) return;
+
+      const chunks = samplesRef.current;
+      if (chunks.length === 0) return;
+
+      samplesRef.current = [];
+      void transcribeChunk(mergeSamples(chunks), generation);
+    },
+    [isActiveGeneration, transcribeChunk]
+  );
+
+  const handleSamples = useCallback(
+    (samples: Float32Array, generation: RecorderGeneration) => {
+      if (!isActiveGeneration(generation)) return;
+
+      const now = Date.now();
+
+      if (rms(samples) > RMS_THRESHOLD) {
+        if (!isSpeakingRef.current) {
+          isSpeakingRef.current = true;
+          speechStartRef.current = now;
+        }
+        silenceStartRef.current = 0;
+        samplesRef.current.push(new Float32Array(samples));
+      } else if (isSpeakingRef.current) {
+        samplesRef.current.push(new Float32Array(samples));
+
+        if (silenceStartRef.current === 0) {
+          silenceStartRef.current = now;
+        } else if (now - silenceStartRef.current > SILENCE_MS) {
+          if (now - speechStartRef.current > MIN_SPEECH_MS) {
+            flush(generation);
+          } else {
+            samplesRef.current = [];
+          }
+          isSpeakingRef.current = false;
+          silenceStartRef.current = 0;
+        }
+      }
+    },
+    [flush, isActiveGeneration]
+  );
+
+  const cancelActiveGeneration = useCallback(() => {
+    const generation = activeGenerationRef.current;
+    activeGenerationRef.current = null;
+    if (generation) cleanupGeneration(generation);
+    resetSpeech();
+
+    if (mountedRef.current) {
+      setIsRecording(false);
+      setIsTranscribing(false);
+    }
+  }, [resetSpeech]);
+
+  const stopRecording = useCallback(() => {
+    const finalChunks = isSpeakingRef.current ? samplesRef.current : [];
+    cancelActiveGeneration();
+
+    if (mountedRef.current && finalChunks.length > 0) {
+      const finalGeneration: RecorderGeneration = {
+        audioContext: null,
+        cancelled: false,
+        completesAfterTranscription: true,
+        pendingTranscriptions: 0,
+        stream: null,
+        worklet: null,
+      };
+      activeGenerationRef.current = finalGeneration;
+      void transcribeChunk(mergeSamples(finalChunks), finalGeneration);
+    }
+  }, [cancelActiveGeneration, transcribeChunk]);
+
   const startRecording = useCallback(async () => {
     if (!isEnabled) {
-      onError('Voice dictation is not enabled');
+      onErrorRef.current('Voice dictation is not enabled');
       return;
     }
 
+    cancelActiveGeneration();
+    const generation: RecorderGeneration = {
+      audioContext: null,
+      cancelled: false,
+      completesAfterTranscription: false,
+      pendingTranscriptions: 0,
+      stream: null,
+      worklet: null,
+    };
+    activeGenerationRef.current = generation;
+
     try {
       const preferredMic = await read('voice_dictation_preferred_mic', false);
+      if (!isActiveGeneration(generation)) return;
 
       const audioConstraints: MediaTrackConstraints = {
         echoCancellation: true,
@@ -215,6 +309,7 @@ export const useAudioRecorder = ({ onTranscription, onError }: UseAudioRecorderO
       try {
         stream = await navigator.mediaDevices.getUserMedia({ audio: audioConstraints });
       } catch (e) {
+        if (!isActiveGeneration(generation)) return;
         if (
           preferredMic &&
           e instanceof DOMException &&
@@ -226,17 +321,26 @@ export const useAudioRecorder = ({ onTranscription, onError }: UseAudioRecorderO
           throw e;
         }
       }
-      streamRef.current = stream;
+      generation.stream = stream;
+      if (!isActiveGeneration(generation)) {
+        cleanupGeneration(generation);
+        return;
+      }
 
       const ctx = new AudioContext({ sampleRate: SAMPLE_RATE });
-      audioContextRef.current = ctx;
+      generation.audioContext = ctx;
 
       await ctx.audioWorklet.addModule(WORKLET_URL);
+      if (!isActiveGeneration(generation)) {
+        cleanupGeneration(generation);
+        return;
+      }
 
       const source = ctx.createMediaStreamSource(stream);
       const worklet = new AudioWorkletNode(ctx, 'audio-capture');
+      generation.worklet = worklet;
 
-      worklet.port.onmessage = (e: MessageEvent<Float32Array>) => handleSamples(e.data);
+      worklet.port.onmessage = (e: MessageEvent<Float32Array>) => handleSamples(e.data, generation);
 
       // Connect through silent gain to keep worklet processing alive
       const silence = ctx.createGain();
@@ -245,19 +349,31 @@ export const useAudioRecorder = ({ onTranscription, onError }: UseAudioRecorderO
       worklet.connect(silence);
       silence.connect(ctx.destination);
 
-      setIsRecording(true);
+      if (isActiveGeneration(generation)) {
+        setIsRecording(true);
+      }
     } catch (error) {
-      stopRecording();
-      onError(errorMessage(error));
+      const isCurrent = isActiveGeneration(generation);
+      if (activeGenerationRef.current === generation) {
+        activeGenerationRef.current = null;
+      }
+      cleanupGeneration(generation);
+      if (isCurrent) {
+        resetSpeech();
+        setIsRecording(false);
+        setIsTranscribing(false);
+        onErrorRef.current(errorMessage(error));
+      }
     }
-  }, [isEnabled, onError, handleSamples, stopRecording, read]);
+  }, [cancelActiveGeneration, handleSamples, isActiveGeneration, isEnabled, read, resetSpeech]);
 
   useEffect(() => {
+    mountedRef.current = true;
     return () => {
-      audioContextRef.current?.close();
-      streamRef.current?.getTracks().forEach((t) => t.stop());
+      mountedRef.current = false;
+      cancelActiveGeneration();
     };
-  }, []);
+  }, [cancelActiveGeneration]);
 
   return {
     isEnabled,