feat(voice): per-account choice between Whisper and the browser; STT_ALLOW_BROWSER

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

frontend/src/lib/__tests__/voice.preference.test.ts

@@ -0,0 +1,41 @@
+import { afterEach, describe, expect, it, vi } from 'vitest';
+
+const getAuthStatus = vi.fn();
+vi.mock('../api', () => ({ api: { getAuthStatus: (...a: any[]) => getAuthStatus(...a) } }));
+
+import { loadVoiceConfig, resetVoiceConfig, setVoicePreference, voiceChoiceAvailable } from '../voice';
+
+// Recording and recognition as far as jsdom needs them to count as available
+(window as any).MediaRecorder = function () {};
+Object.defineProperty(navigator, 'mediaDevices', { value: { getUserMedia: vi.fn() }, configurable: true });
+(window as any).webkitSpeechRecognition = function () {};
+
+afterEach(() => {
+  resetVoiceConfig();
+  getAuthStatus.mockReset();
+});
+
+describe('voice input: Whisper or browser per account', () => {
+  it('uses the browser when the account picked it and the server allows it', async () => {
+    getAuthStatus.mockResolvedValue({ voice: { mode: 'server', maxSeconds: 600, browserAllowed: true } });
+    expect((await loadVoiceConfig()).mode).toBe('server');
+    expect(voiceChoiceAvailable()).toBe(true);
+    setVoicePreference('browser');
+    expect((await loadVoiceConfig()).mode).toBe('browser');
+    setVoicePreference('');
+    expect((await loadVoiceConfig()).mode).toBe('server');
+  });
+
+  it('keeps Whisper when the server forbids the browser (STT_ALLOW_BROWSER=false)', async () => {
+    getAuthStatus.mockResolvedValue({ voice: { mode: 'server', maxSeconds: 600, browserAllowed: false } });
+    setVoicePreference('browser');
+    expect((await loadVoiceConfig()).mode).toBe('server');
+    expect(voiceChoiceAvailable()).toBe(false);
+  });
+
+  it('offers no choice without a Whisper server', async () => {
+    getAuthStatus.mockResolvedValue({ voice: { mode: 'browser', maxSeconds: 600, browserAllowed: true } });
+    await loadVoiceConfig();
+    expect(voiceChoiceAvailable()).toBe(false);
+  });
+});

frontend/src/lib/api.ts

@@ -301,6 +301,7 @@ export const api = {
     notifyBeforeMinutes?: number;
     notificationSettings?: NotificationSettings;
     doneRetentionDays?: number;
+    voiceInput?: '' | 'browser';
   }): Promise<User> {
     return request<User>('/api/v1/auth/settings', {
       method: 'PATCH',

frontend/src/lib/types.ts

@@ -31,7 +31,7 @@ export interface AuthStatus {
     customLinkUrl?: string;
   };
   // Voice input: own Whisper server, the browser's recognition, or switched off (STT_* environment)
-  voice?: { mode: VoiceMode; maxSeconds: number };
+  voice?: { mode: VoiceMode; maxSeconds: number; browserAllowed?: boolean };
 }
 
 export type VoiceMode = 'server' | 'browser' | 'off';
@@ -113,6 +113,8 @@ export interface User {
   notificationSettings?: NotificationSettings;
   // Completed tasks are deleted this many days after ticking them off (0 = never)
   doneRetentionDays?: number;
+  // Voice input: '' = the server's Whisper (default), 'browser' = the browser's recognition
+  voiceInput?: '' | 'browser';
   organizationId?: string;
   createdAt: string;
 }

frontend/src/lib/useVoiceInput.ts

@@ -1,6 +1,6 @@
 import { useCallback, useEffect, useRef, useState } from 'react';
 import { api } from './api';
-import { browserRecognition, currentVoiceConfig, EffectiveVoiceMode, joinSpeechResults, loadVoiceConfig, VoiceRecorder } from './voice';
+import { browserRecognition, currentVoiceConfig, EffectiveVoiceMode, joinSpeechResults, loadVoiceConfig, VOICE_CONFIG_EVENT, VoiceRecorder } from './voice';
 
 export type VoiceState = 'idle' | 'listening' | 'transcribing';
 
@@ -63,12 +63,17 @@ export function useVoiceInput({ onText, onError, continuous = false, live = fals
 
   useEffect(() => {
     mountedRef.current = true;
-    void loadVoiceConfig().then((cfg) => {
-      if (!mountedRef.current) return;
-      setMode(cfg.mode);
-      maxSecondsRef.current = cfg.maxSeconds;
-    });
+    const load = () =>
+      void loadVoiceConfig().then((cfg) => {
+        if (!mountedRef.current) return;
+        setMode(cfg.mode);
+        maxSecondsRef.current = cfg.maxSeconds;
+      });
+    load();
+    // The account setting (Whisper or browser) can change while this button is on screen
+    window.addEventListener(VOICE_CONFIG_EVENT, load);
     return () => {
+      window.removeEventListener(VOICE_CONFIG_EVENT, load);
       mountedRef.current = false;
       clearAllTimers();
       recorderRef.current?.cancel();

frontend/src/lib/voice.ts

@@ -3,6 +3,8 @@
 //   server   record with MediaRecorder, the own Whisper server turns it into text after stopping
 //   browser  the browser's speech recognition (Chrome, Edge, Safari; text appears while speaking)
 //   off      no voice input
+// With a Whisper server everyone may pick the browser instead (account setting), unless the server
+// forbids it (STT_ALLOW_BROWSER=false).
 import { api } from './api';
 import type { VoiceMode } from './types';
 
@@ -15,6 +17,33 @@ export interface VoiceConfig {
 
 let configPromise: Promise<VoiceConfig> | null = null;
 let loadedConfig: VoiceConfig | null = null;
+let serverMode: VoiceMode | undefined;
+let browserAllowed = true;
+let preference: '' | 'browser' = '';
+
+/** Fired when the effective mode changes (the account setting was loaded or changed). */
+export const VOICE_CONFIG_EVENT = 'mindsift:voice-config';
+
+function applyPreference(mode: VoiceMode | undefined): VoiceMode | undefined {
+  return mode === 'server' && browserAllowed && preference === 'browser' ? 'browser' : mode;
+}
+
+/** The account's choice ('' = Whisper server, 'browser'); set when the user is loaded or changes it. */
+export function setVoicePreference(next: '' | 'browser' | undefined): void {
+  const value = next === 'browser' ? 'browser' : '';
+  if (value === preference) return;
+  preference = value;
+  if (loadedConfig) {
+    loadedConfig = { ...loadedConfig, mode: effectiveMode(applyPreference(serverMode)) };
+    configPromise = Promise.resolve(loadedConfig);
+  }
+  if (typeof window !== 'undefined') window.dispatchEvent(new CustomEvent(VOICE_CONFIG_EVENT));
+}
+
+/** Whether the account may choose between the Whisper server and the browser. */
+export function voiceChoiceAvailable(): boolean {
+  return serverMode === 'server' && browserAllowed;
+}
 
 export function browserRecognition(): any {
   if (typeof window === 'undefined') return null;
@@ -38,7 +67,9 @@ export function loadVoiceConfig(): Promise<VoiceConfig> {
     configPromise = api
       .getAuthStatus()
       .then((st) => {
-        loadedConfig = { mode: effectiveMode(st.voice?.mode), maxSeconds: st.voice?.maxSeconds || 600 };
+        serverMode = st.voice?.mode;
+        browserAllowed = st.voice?.browserAllowed !== false;
+        loadedConfig = { mode: effectiveMode(applyPreference(serverMode)), maxSeconds: st.voice?.maxSeconds || 600 };
         return loadedConfig;
       })
       .catch(() => {
@@ -58,6 +89,9 @@ export function currentVoiceConfig(): VoiceConfig {
 export function resetVoiceConfig(): void {
   configPromise = null;
   loadedConfig = null;
+  serverMode = undefined;
+  browserAllowed = true;
+  preference = '';
 }
 
 // Containers in order of preference; Safari only records MP4/AAC