Skip to main content

Set Up Text-to-Speech Voices

Text-to-speech turns typed cards into audio without you ever pressing record. It is the fastest way to build a deck, the easiest way to keep hundreds of cards consistent, and often the only way to study a language you cannot pronounce yet.

Everything about voices lives in Settings > Text-To-Speech, which has two screens:

  • My Voices — the voices you have set up, and which one is the app default.
  • Card Side Voices — which voice reads each side of a new card.

My Voices​

My Voices is your shortlist. Each entry is a voice with its own name and settings. The selected voice (the one with the tick) is the app default — used whenever nothing more specific applies. Tap a voice to make it the default, the play button to hear it, and swipe or long press to Edit or Delete it.

The app starts you with one voice from your device, so text-to-speech works from the first card.

Tap + to add a voice, and choose a source:

SourceSoundNeedsWorks offline
System VoiceVaries; Enhanced and Premium voices are good.Nothing — uses voices installed on your device.Yes
Kokoro (on-device)Natural neural speech.A one-off model download (86 MB and up).Yes, once downloaded
Amazon PollyWide language and accent coverage.Your own AWS credentials.No
OpenAIVery natural, especially in English.Your own OpenAI API key.No

You can keep voices from different sources side by side and use them on the same card.

System Voice​

Pick a Language, then a voice from the list — search helps when there are many — and adjust Rate and Pitch. Reset to Default undoes your adjustments. Give it a Name you will recognise and tap Save.

The list only shows voices installed on your device. See Get more system voices to add higher-quality ones.

Kokoro (on-device)​

Kokoro is a neural model that runs entirely on your phone: no account, no key, no per-character cost, and no connection once downloaded.

  1. Under Model, choose a Quality and tap Download Model. The model is shared by every Kokoro voice you create, so you only do this once.

    QualityDownload
    Smallest (q8f16)86 MB
    Balanced (int8)88 MB
    High quality (fp16)169 MB
    Full quality (fp32)326 MB

    Smallest is the default on iPhone and iPad, and the right choice for almost everyone. Move up only if you can hear the difference on your own cards. Android defaults to Balanced, because the smallest build does not run on Android.

  2. Choose a Language: English (American), English (British), German, Spanish, French, Hindi, Italian, Japanese, Portuguese (Brazil), or Mandarin Chinese.

  3. Tap a Voice to download it — each is a small file — then select it.

  4. Name it, try it with Preview Voice, and Save.

A few voices use their own dedicated model instead of the shared one; the Model section says so when that applies. Generating audio on-device is slower than fetching it from the cloud, so a large CSV import takes a while.

Amazon Polly and OpenAI​

Both are bring your own account: you paste in your own credentials and pay the provider directly for what you use. There is no markup.

  • Amazon Polly — enter an Access key ID and Secret access key, choose an Engine and Region, then Fetch Voices and pick a Language and Voice.
  • OpenAI — enter your API Key, choose a Model, and pick a Voice. Recommended voices are marked.

New to AWS? Create Amazon Polly access keys walks through making a user and keys for Polly.

Credentials are stored only in this device's Keychain. They do not sync, so on another device the voice appears but asks you to re-create it with your credentials before it can generate audio.

Card Side Voices​

Card Side Voices saves you choosing a voice every time. It assigns a default voice to each side position — one for every Question, another for every Answer — and is used by:

  • the text-to-speech composer when you add a new side to a card, and
  • CSV import, which reads every side aloud with its assigned voice.

Open Settings > Text-To-Speech > Card Side Voices, choose a voice for each side, and preview it with the play button. Add Side extends the list for cards with three or more sides; Remove Last Side trims it. A side without an assignment uses the app default.

This matters most for language learning, where questions and answers are in different languages. It only sets the starting point — you can pick any voice on an individual side.

Change the voice on an existing side​

Open the card, tap ⋯ beside the side, and choose Edit. Change the text or the voice, tap Regenerate Audio, and keep it with the checkmark.

When a voice goes missing​

System voices can disappear — you deleted one from your device, moved to a new phone, or synced to a device that never had it. Audio already generated keeps playing: it is a saved file, not a live render. The voice is marked Unavailable on this device, and a side that used it falls back to your app default the next time you edit it. Download the voice again, or pick a replacement.

Get more system voices​

iPhone and iPad​

The higher-quality Enhanced and Premium voices are optional downloads.

  1. Open the Settings app.
  2. Go to Accessibility > Spoken Content > Voices. On recent versions of iOS this may be under Accessibility > Read & Speak.
  3. Choose a language and tap the download button beside a voice.
  4. Return to Audio Flashcards and add it under My Voices > + > System Voice.

Personal Voice and Siri voices are not available in Audio Flashcards.

Android​

  1. Open Settings and search for Text-to-speech output. It may be under Accessibility, System > Languages & input, or General management, depending on your phone.
  2. Tap the gear beside your engine (usually Google or Samsung) to download languages and voices.
  3. Restart Audio Flashcards so the new voices appear.

If your engine offers a network or high-quality tier, turn it on — the difference is usually significant.