feat(voice): live dictation with browser pause restart and server audio chunking

frontend/src/lib/__tests__/useVoiceInput.test.tsx

@@ -130,4 +130,121 @@ describe('useVoiceInput', () => {
     await waitFor(() => expect(none.result.current.mode).toBe('unsupported'));
     expect(none.result.current.available).toBe(false);
   });
+
+  it('browser mode with live=true: restarts on pause (onend), keeps base text, rate-limits restarts to 300ms, and sends final=true only on stop()', async () => {
+    const instances: any[] = [];
+    class FakeSpeechRecognition {
+      lang = '';
+      interimResults = false;
+      continuous = false;
+      onresult: ((e: any) => void) | null = null;
+      onerror: ((e: any) => void) | null = null;
+      onend: (() => void) | null = null;
+      constructor() {
+        instances.push(this);
+      }
+      start() {}
+      stop() {
+        this.onend?.();
+      }
+      abort() {}
+    }
+    (window as any).SpeechRecognition = FakeSpeechRecognition;
+    getAuthStatus.mockResolvedValue({ voice: { mode: 'browser', maxSeconds: 600 } });
+
+    const onText = vi.fn();
+    const { result } = renderHook(() => useVoiceInput({ onText, live: true }));
+    await waitFor(() => expect(result.current.mode).toBe('browser'));
+
+    vi.useFakeTimers();
+    await act(async () => {
+      await result.current.start();
+    });
+    expect(instances).toHaveLength(1);
+
+    // First utterance before a pause
+    act(() => {
+      instances[0].onresult?.({ results: [[{ transcript: 'Erster Satz.' }]] });
+    });
+    expect(onText).toHaveBeenLastCalledWith('Erster Satz.', false);
+
+    // Browser ends recognition on speech pause -> immediate restart (#2)
+    act(() => {
+      instances[0].onend?.();
+    });
+    expect(instances).toHaveLength(2);
+    expect(onText).not.toHaveBeenCalledWith(expect.anything(), true);
+
+    // Another immediate onend within <300ms waits until 300ms have passed
+    act(() => {
+      instances[1].onend?.();
+    });
+    expect(instances).toHaveLength(2);
+    act(() => {
+      vi.advanceTimersByTime(300);
+    });
+    expect(instances).toHaveLength(3);
+
+    // Second utterance appends to base text
+    act(() => {
+      instances[2].onresult?.({ results: [[{ transcript: 'Zweiter Satz.' }]] });
+    });
+    expect(onText).toHaveBeenLastCalledWith('Erster Satz. Zweiter Satz.', false);
+
+    // Explicit stop() finishes with final=true
+    await act(async () => {
+      await result.current.stop();
+    });
+    expect(onText).toHaveBeenLastCalledWith('Erster Satz. Zweiter Satz.', true);
+    expect(result.current.state).toBe('idle');
+  });
+
+  it('server mode with live=true: cuts segments onto new MediaRecorders on the same stream, orders out-of-order responses by chunk index, and releases stream at the end', async () => {
+    getAuthStatus.mockResolvedValue({ voice: { mode: 'server', maxSeconds: 600 } });
+
+    const resolvers: Array<(text: string) => void> = [];
+    transcribe.mockImplementation(
+      () =>
+        new Promise<string>((resolve) => {
+          resolvers.push(resolve);
+        })
+    );
+
+    const onText = vi.fn();
+    const { result } = renderHook(() => useVoiceInput({ onText, live: true }));
+    await waitFor(() => expect(result.current.mode).toBe('server'));
+
+    vi.useFakeTimers();
+    await act(async () => {
+      await result.current.start();
+    });
+
+    // Advance 8s -> segment 0 is cut and sent to transcribe, stream is NOT released yet
+    await act(async () => {
+      await vi.advanceTimersByTimeAsync(8000);
+    });
+    expect(transcribe).toHaveBeenCalledTimes(1);
+    expect(stopTrack).not.toHaveBeenCalled();
+
+    // Advance 4s and stop -> segment 1 (final) is sent to transcribe, stream IS released now
+    await act(async () => {
+      await vi.advanceTimersByTimeAsync(4000);
+      await result.current.stop();
+    });
+    expect(transcribe).toHaveBeenCalledTimes(2);
+    expect(stopTrack).toHaveBeenCalledTimes(1);
+
+    // Resolve chunk 1 BEFORE chunk 0 to verify index-based ordering
+    await act(async () => {
+      resolvers[1]('zweiter Teil');
+    });
+    expect(onText).toHaveBeenLastCalledWith('zweiter Teil', false);
+
+    // Now resolve chunk 0 -> both chunks are done, ordered "Erster Teil zweiter Teil" with final=true
+    await act(async () => {
+      resolvers[0]('Erster Teil');
+    });
+    expect(onText).toHaveBeenLastCalledWith('Erster Teil zweiter Teil', true);
+    expect(result.current.state).toBe('idle');
+  });
 });