🌩️ feat: cloud-based browser voices (#3297)

* initial voice support

* feat: local voices; feat: switch cloud-based voices

* feat: apply voice to hook

client/src/components/Nav/SettingsTabs/Speech/Speech.tsx

@@ -8,20 +8,21 @@ import store from '~/store';
 import { cn } from '~/utils';
 import ConversationModeSwitch from './ConversationModeSwitch';
 import {
+  CloudBrowserVoicesSwitch,
+  AutomaticPlaybackSwitch,
   TextToSpeechSwitch,
   EngineTTSDropdown,
-  AutomaticPlaybackSwitch,
   CacheTTSSwitch,
   VoiceDropdown,
   PlaybackRate,
 } from './TTS';
 import {
-  DecibelSelector,
+  AutoTranscribeAudioSwitch,
-  EngineSTTDropdown,
   LanguageSTTDropdown,
   SpeechToTextSwitch,
   AutoSendTextSwitch,
-  AutoTranscribeAudioSwitch,
+  EngineSTTDropdown,
+  DecibelSelector,
 } from './STT';
 import { useGetCustomConfigSpeechQuery } from 'librechat-data-provider/react-query';
 
@@ -42,6 +43,9 @@ function Speech() {
   const [autoSendText, setAutoSendText] = useRecoilState(store.autoSendText);
   const [engineTTS, setEngineTTS] = useRecoilState<string>(store.engineTTS);
   const [voice, setVoice] = useRecoilState<string>(store.voice);
+  const [cloudBrowserVoices, setCloudBrowserVoices] = useRecoilState<boolean>(
+    store.cloudBrowserVoices,
+  );
   const [languageTTS, setLanguageTTS] = useRecoilState<string>(store.languageTTS);
   const [automaticPlayback, setAutomaticPlayback] = useRecoilState(store.automaticPlayback);
   const [playbackRate, setPlaybackRate] = useRecoilState(store.playbackRate);
@@ -61,6 +65,7 @@ function Speech() {
         autoSendText: { value: autoSendText, setFunc: setAutoSendText },
         engineTTS: { value: engineTTS, setFunc: setEngineTTS },
         voice: { value: voice, setFunc: setVoice },
+        cloudBrowserVoices: { value: cloudBrowserVoices, setFunc: setCloudBrowserVoices },
         languageTTS: { value: languageTTS, setFunc: setLanguageTTS },
         automaticPlayback: { value: automaticPlayback, setFunc: setAutomaticPlayback },
         playbackRate: { value: playbackRate, setFunc: setPlaybackRate },
@@ -86,6 +91,7 @@ function Speech() {
       autoSendText,
       engineTTS,
       voice,
+      cloudBrowserVoices,
       languageTTS,
       automaticPlayback,
       playbackRate,
@@ -101,6 +107,7 @@ function Speech() {
       setAutoSendText,
       setEngineTTS,
       setVoice,
+      setCloudBrowserVoices,
       setLanguageTTS,
       setAutomaticPlayback,
       setPlaybackRate,
@@ -168,27 +175,23 @@ function Speech() {
 
         <Tabs.Content value={'simple'}>
           <div className="flex flex-col gap-3 text-sm text-black dark:text-gray-50">
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
-              <ConversationModeSwitch />
-            </div>
-            <div className="h-px bg-black/20 bg-white/20" role="none" />
             <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
               <SpeechToTextSwitch />
             </div>
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            <div className="border-b last-of-type:border-b-0 dark:border-gray-700">
               <EngineSTTDropdown />
             </div>
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            <div className="border-b last-of-type:border-b-0 dark:border-gray-700">
               <LanguageSTTDropdown />
             </div>
             <div className="h-px bg-black/20 bg-white/20" role="none" />
             <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
               <TextToSpeechSwitch />
             </div>
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            <div className="border-b last-of-type:border-b-0 dark:border-gray-700">
               <EngineTTSDropdown />
             </div>
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            <div className="border-b last-of-type:border-b-0 dark:border-gray-700">
               <VoiceDropdown />
             </div>
           </div>
@@ -196,47 +199,52 @@ function Speech() {
 
         <Tabs.Content value={'advanced'}>
           <div className="flex flex-col gap-3 text-sm text-black dark:text-gray-50">
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            <div className="border-b last-of-type:border-b-0 dark:border-gray-700">
               <ConversationModeSwitch />
             </div>
             <div className="h-px bg-black/20 bg-white/20" role="none" />
             <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
               <SpeechToTextSwitch />
             </div>
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            <div className="border-b last-of-type:border-b-0 dark:border-gray-700">
               <EngineSTTDropdown />
             </div>
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            <div className="border-b last-of-type:border-b-0 dark:border-gray-700">
               <LanguageSTTDropdown />
             </div>
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            <div className="border-b pb-2 last-of-type:border-b-0 dark:border-gray-700">
               <AutoTranscribeAudioSwitch />
             </div>
             {autoTranscribeAudio && (
-              <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+              <div className="border-b pb-2 last-of-type:border-b-0 dark:border-gray-700">
                 <DecibelSelector />
               </div>
             )}
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            <div className="border-b last-of-type:border-b-0 dark:border-gray-700">
               <AutoSendTextSwitch />
             </div>
             <div className="h-px bg-black/20 bg-white/20" role="none" />
             <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
               <TextToSpeechSwitch />
             </div>
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            <div className="border-b last-of-type:border-b-0 dark:border-gray-700">
               <AutomaticPlaybackSwitch />
             </div>
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            <div className="border-b last-of-type:border-b-0 dark:border-gray-700">
               <EngineTTSDropdown />
             </div>
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            <div className="border-b last-of-type:border-b-0 dark:border-gray-700">
               <VoiceDropdown />
             </div>
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            {engineTTS === 'browser' && (
+              <div className="border-b pb-2 last-of-type:border-b-0 dark:border-gray-700">
+                <CloudBrowserVoicesSwitch />
+              </div>
+            )}
+            <div className="border-b pb-2 last-of-type:border-b-0 dark:border-gray-700">
               <PlaybackRate />
             </div>
-            <div className="border-b pb-3 last-of-type:border-b-0 dark:border-gray-700">
+            <div className="border-b last-of-type:border-b-0 dark:border-gray-700">
               <CacheTTSSwitch />
             </div>
           </div>

client/src/components/Nav/SettingsTabs/Speech/TTS/CloudBrowserVoicesSwitch.tsx

@@ -0,0 +1,37 @@
+import { useRecoilState } from 'recoil';
+import { Switch } from '~/components/ui';
+import { useLocalize } from '~/hooks';
+import store from '~/store';
+
+export default function CloudBrowserVoicesSwitch({
+  onCheckedChange,
+}: {
+  onCheckedChange?: (value: boolean) => void;
+}) {
+  const localize = useLocalize();
+  const [cloudBrowserVoices, setCloudBrowserVoices] = useRecoilState<boolean>(
+    store.cloudBrowserVoices,
+  );
+  const [textToSpeech] = useRecoilState<boolean>(store.textToSpeech);
+
+  const handleCheckedChange = (value: boolean) => {
+    setCloudBrowserVoices(value);
+    if (onCheckedChange) {
+      onCheckedChange(value);
+    }
+  };
+
+  return (
+    <div className="flex items-center justify-between">
+      <div>{localize('com_nav_enable_cloud_browser_voice')}</div>
+      <Switch
+        id="CloudBrowserVoices"
+        checked={cloudBrowserVoices}
+        onCheckedChange={handleCheckedChange}
+        className="ml-4"
+        data-testid="CloudBrowserVoices"
+        disabled={!textToSpeech}
+      />
+    </div>
+  );
+}

client/src/components/Nav/SettingsTabs/Speech/TTS/VoiceDropdown.tsx

@@ -1,34 +1,73 @@
+import React, { useMemo, useEffect, useState } from 'react';
 import { useRecoilState } from 'recoil';
-import { useMemo, useEffect } from 'react';
 import Dropdown from '~/components/ui/DropdownNoState';
 import { useVoicesQuery } from '~/data-provider';
 import { useLocalize } from '~/hooks';
 import store from '~/store';
 
+const getLocalVoices = (): Promise<SpeechSynthesisVoice[]> => {
+  return new Promise((resolve) => {
+    const voices = speechSynthesis.getVoices();
+    console.log('voices', voices);
+
+    if (voices.length) {
+      resolve(voices);
+    } else {
+      speechSynthesis.onvoiceschanged = () => resolve(speechSynthesis.getVoices());
+    }
+  });
+};
+
+type VoiceOption = {
+  value: string;
+  display: string;
+};
+
 export default function VoiceDropdown() {
   const localize = useLocalize();
   const [voice, setVoice] = useRecoilState(store.voice);
-  const { data } = useVoicesQuery();
+  const [engineTTS] = useRecoilState(store.engineTTS);
+  const [cloudBrowserVoices] = useRecoilState(store.cloudBrowserVoices);
+  const externalTextToSpeech = engineTTS === 'external';
+  const { data: externalVoices = [] } = useVoicesQuery();
+  const [localVoices, setLocalVoices] = useState<SpeechSynthesisVoice[]>([]);
 
   useEffect(() => {
-    if (!voice && data?.length) {
+    if (!externalTextToSpeech) {
-      setVoice(data[0]);
+      getLocalVoices().then(setLocalVoices);
     }
-  }, [voice, data, setVoice]);
+  }, [externalTextToSpeech]);
 
-  const voiceOptions = useMemo(
+  useEffect(() => {
-    () => (data ?? []).map((v: string) => ({ value: v, display: v })),
+    if (voice) {
-    [data],
+      return;
-  );
+    }
+
+    if (externalTextToSpeech && externalVoices.length) {
+      setVoice(externalVoices[0]);
+    } else if (!externalTextToSpeech && localVoices.length) {
+      setVoice(localVoices[0].name);
+    }
+  }, [voice, setVoice, externalTextToSpeech, externalVoices, localVoices]);
+
+  const voiceOptions: VoiceOption[] = useMemo(() => {
+    if (externalTextToSpeech) {
+      return externalVoices.map((v) => ({ value: v, display: v }));
+    } else {
+      return localVoices
+        .filter((v) => cloudBrowserVoices || v.localService === true)
+        .map((v) => ({ value: v.name, display: v.name }));
+    }
+  }, [externalTextToSpeech, externalVoices, localVoices, cloudBrowserVoices]);
 
   return (
     <div className="flex items-center justify-between">
       <div>{localize('com_nav_voice_select')}</div>
       <Dropdown
         value={voice}
-        onChange={(value: string) => setVoice(value)}
+        onChange={setVoice}
         options={voiceOptions}
-        position={'left'}
+        position="left"
         testId="VoiceDropdown"
       />
     </div>

client/src/components/Nav/SettingsTabs/Speech/TTS/__tests__/CloudBrowserVoicesSwitch.spec.tsx

@@ -0,0 +1,41 @@
+import React from 'react';
+import '@testing-library/jest-dom/extend-expect';
+import { render, fireEvent } from 'test/layout-test-utils';
+import CloudBrowserVoicesSwitch from '../CloudBrowserVoicesSwitch';
+import { RecoilRoot } from 'recoil';
+
+describe('CloudBrowserVoicesSwitch', () => {
+  /**
+   * Mock function to set the cache-tts state.
+   */
+  let mockSetCloudBrowserVoices:
+    | jest.Mock<void, [boolean]>
+    | ((value: boolean) => void)
+    | undefined;
+
+  beforeEach(() => {
+    mockSetCloudBrowserVoices = jest.fn();
+  });
+
+  it('renders correctly', () => {
+    const { getByTestId } = render(
+      <RecoilRoot>
+        <CloudBrowserVoicesSwitch />
+      </RecoilRoot>,
+    );
+
+    expect(getByTestId('CloudBrowserVoices')).toBeInTheDocument();
+  });
+
+  it('calls onCheckedChange when the switch is toggled', () => {
+    const { getByTestId } = render(
+      <RecoilRoot>
+        <CloudBrowserVoicesSwitch onCheckedChange={mockSetCloudBrowserVoices} />
+      </RecoilRoot>,
+    );
+    const switchElement = getByTestId('CloudBrowserVoices');
+    fireEvent.click(switchElement);
+
+    expect(mockSetCloudBrowserVoices).toHaveBeenCalledWith(true);
+  });
+});

client/src/components/Nav/SettingsTabs/Speech/TTS/index.ts

@@ -1,6 +1,7 @@
+export { default as CloudBrowserVoicesSwitch } from './CloudBrowserVoicesSwitch';
 export { default as AutomaticPlaybackSwitch } from './AutomaticPlaybackSwitch';
-export { default as CacheTTSSwitch } from './CacheTTSSwitch';
-export { default as EngineTTSDropdown } from './EngineTTSDropdown';
-export { default as PlaybackRate } from './PlaybackRate';
 export { default as TextToSpeechSwitch } from './TextToSpeechSwitch';
+export { default as EngineTTSDropdown } from './EngineTTSDropdown';
+export { default as CacheTTSSwitch } from './CacheTTSSwitch';
 export { default as VoiceDropdown } from './VoiceDropdown';
+export { default as PlaybackRate } from './PlaybackRate';

client/src/hooks/Input/useTextToSpeechBrowser.ts

@@ -1,12 +1,24 @@
+import { useRecoilState } from 'recoil';
 import { useState } from 'react';
+import store from '~/store';
 
 function useTextToSpeechBrowser() {
+  const [cloudBrowserVoices] = useRecoilState(store.cloudBrowserVoices);
   const [isSpeaking, setIsSpeaking] = useState(false);
+  const [voiceName] = useRecoilState(store.voice);
 
   const generateSpeechLocal = (text: string) => {
     const synth = window.speechSynthesis;
+    const voices = synth.getVoices().filter((v) => cloudBrowserVoices || v.localService === true);
+    const voice = voices.find((v) => v.name === voiceName);
+
+    if (!voice) {
+      return;
+    }
+
     synth.cancel();
     const utterance = new SpeechSynthesisUtterance(text);
+    utterance.voice = voice;
     utterance.onend = () => {
       setIsSpeaking(false);
     };

client/src/localization/languages/Eng.ts

@@ -641,6 +641,7 @@ export default {
   com_nav_delete_cache_storage: 'Delete TTS cache storage',
   com_nav_enable_cache_tts: 'Enable cache TTS',
   com_nav_voice_select: 'Voice',
+  com_nav_enable_cloud_browser_voice: 'Use cloud-based voices',
   com_nav_info_enter_to_send:
     'When enabled, pressing `ENTER` will send your message. When disabled, pressing Enter will add a new line, and you\'ll need to press `CTRL + ENTER` to send your message.',
   com_nav_info_save_draft:

client/src/store/settings.ts

@@ -50,6 +50,7 @@ const localStorageAtoms = {
   textToSpeech: atomWithLocalStorage('textToSpeech', true),
   engineTTS: atomWithLocalStorage('engineTTS', 'browser'),
   voice: atomWithLocalStorage('voice', ''),
+  cloudBrowserVoices: atomWithLocalStorage('cloudBrowserVoices', false),
   languageTTS: atomWithLocalStorage('languageTTS', ''),
   automaticPlayback: atomWithLocalStorage('automaticPlayback', false),
   playbackRate: atomWithLocalStorage<number | null>('playbackRate', null),