Kokoro TTS · On-device
Kokoro TTS, running entirely on your iPhone
Sandbook is a free Kokoro TTS app that runs on iPhone, iPad and Mac. The whole Kokoro-82M model — all 82 million parameters — is bundled in the download with all 41 voices, and runs on your device. There is no server call, no API key and no account. 41 voices in English, Spanish, French, Italian, Brazilian Portuguese and Hindi, 24 kHz output.
iOS 18.2 or later · iPhone XS / XR or later · iPad with A12 or later · Mac with M1 or later, macOS 15.2 · No account

In short
Is there a Kokoro TTS app for iPhone?
Yes. Sandbook runs Kokoro-82M on iPhone and iPad (and natively on Macs with Apple silicon), with all 82 million parameters bundled in the download with all 41 voices and executed on the device. It is free, with no in-app purchase, no account and no character limit, with 41 voices in six languages, keeps reading with the screen locked, and needs iOS 18.2 on an iPhone XS, iPhone XR or later. The main limit: voices cover six languages, not dozens.
Last verified November 7, 2026
What a Kokoro TTS app looks like when the model is local
Most ways to hear Kokoro involve a machine that isn't yours: a Python script on a laptop, a Space on someone's GPU, a hosted endpoint behind a key. In Sandbook the model file sits in the app bundle and the inference runs on the phone itself, so nothing is queued, metered or rate-limited and there is no character cap. A 400-page book opens in airplane mode exactly as it does on Wi-Fi.
On a computer, you can try Kokoro in your browser before installing: the same model and all 41 voices run in the tab, free, and nothing you type is uploaded.
Kokoro TTS on iOS, offline
Sandbook is a text to speech app that brings Kokoro TTS to iOS with no server in the loop. The model and all 41 voices arrive with the install: turn on airplane mode and books, PDFs, scanned pages and pasted text still play, with no account, no API key and no upload.
- iPhone and iPad: iOS 18.2 or later on an iPhone XS, iPhone XR or later, or an iPad with an A12 chip or later; it does not run on iOS 17. It keeps reading with the screen locked, with nothing to prepare first.
- Is there a Kokoro TTS app for Mac? Yes: Sandbook for Mac runs the same model and voices natively on Apple silicon (macOS 15.2 or later), offline.
Kokoro-82M on Core ML, on iPhone, iPad and Mac
The engine is Kokoro-82M, the open-weight TTS model, run as a Core ML model on the CPU on every device; that is what lets it keep reading behind the lock. Speech is made sentence by sentence ahead of what you hear, so a chapter begins in about the time it takes to read the chapter title. Speed is 0.5× to 2.5×, applied inside the model: 2× stays a person talking quickly rather than a chipmunk.
How the engine works: the model, the pipeline, the cache
Kokoro-82M is an 82-million-parameter network with an ALBERT-style text encoder, an LSTM duration predictor, a prosody predictor and an iSTFTNet vocoder. Sandbook runs it as a Core ML model on the CPU, on iPhone and iPad since October 2026 and on the Mac since November 2026, in the foreground and with the screen locked alike, because iOS does not let an app use the GPU in the background. Before that, both apps ran it on Apple MLX, Apple's array framework, on the GPU via Metal. The voices are the same 41 on every device.
Text is phonemized, chunked on sentence boundaries, and synthesized sentence by sentence. Playback starts as soon as the first sentence is ready while the next ones generate behind it — you are never waiting for a whole document to render before anything plays.
Speed is applied by scaling predicted phoneme durations, not by resampling the finished waveform. A passage you have already heard, in the same voice and at the same speed, plays back instantly from the device instead of being generated again. In English, very long sentences are read in full.
How fast is Kokoro on an iPhone?
On an iPhone 15, speech started about 2.5 seconds after pressing play on a fresh chapter, and paragraphs then generated 3–6× faster than real time. These timings come from one reading session in October 2026, taken from the app's own log, with the earlier MLX engine, before the iPhone app moved to Core ML; they are not a benchmark suite, and no figure has been published yet for the Core ML engine the apps use now. Older supported iPhones, back to the iPhone XS and XR, are slower; on every one of them, playback starts after the first sentence while the rest renders behind it.
The measurements, paragraph by paragraph and stage by stage
The first paragraph, 13.8 seconds of audio, was generated in 2.4 seconds. After that: 2.9 seconds of audio in 0.9 seconds, 13.6 seconds in 2.4 and 2.2 seconds in 0.6. The model stayed resident at about 220 MB of memory while reading, and a short warm-up at launch, about a second of synthesis, is there so the first sentence comes out quickly.
| Stage | Measured on iPhone 15 |
|---|---|
| Phonemizing | about 0.01 s |
| Text encoder | 0.05–0.09 s |
| Duration predictor | 0.2–0.5 s |
| Prosody predictor | 0.2–0.5 s |
| Decoder (vocoder) | 0.3–1.3 s |
What ships in the download
Sandbook is a large download on purpose: the model and all 41 voices are included, loaded from the bundle rather than fetched. Nothing is downloaded after install, and nothing is streamed at runtime.
- kokoro-v1_0 weights — 82 million parameters, as a Core ML model on iPhone, iPad and Mac.
- 41 voice embeddings, one style tensor per voice, all bundled, none locked or purchased.
- An eSpeak NG build for phonemization, compiled into the app as a framework.
Where Sandbook differs from upstream Kokoro
- Six languages, not all of them. Upstream Kokoro ships voices in several languages. Sandbook includes
en-usanden-gb(28 voices), plus Spanish, French, Italian, Brazilian Portuguese and Hindi, 41 Kokoro voices in all, on an iPhone, offline. eSpeak NG switches language with the voice you pick, and the English text clean-up (dates, prices, units, acronyms) runs only for English voices. - No voice mixing and no custom style vectors. One quality tier, the same 82M model at 24 kHz for every voice. What you can change is how a name or word is said, for one book or for every book.
Getting text into it
- Files: EPUB, PDF, Word, TXT, RTF, Markdown or HTML, through the Files picker, “Open with Sandbook”, or the share sheet from any app.
- Web pages: add a web page with Paste a link, or with Save to Sandbook in the share sheet of Safari, Chrome or another app; the app keeps the article and drops the menus, ads and cookie banners.
- Paper and pictures: Scan printed pages with the camera, or pick a photo, and the text is recognized on the device.
- Your own text: type, paste or read your clipboard.

Books get word-by-word highlighting inside a softly tinted current paragraph, a chapter picker and tap-a-paragraph-to-read, and open at the paragraph where you stopped, each with its own voice and speed. A chapter or the whole book exports as one M4A file with chapter markers, the cover and the title, and so does text in Write text.
Kokoro TTS questions
Is this the same Kokoro as the open-source model?
Yes. Sandbook runs Kokoro-82M — the published open-weight model, same architecture, same 82 million parameters — converted to run on the device with Core ML on the CPU, the same engine on iPhone, iPad and Mac. It is not a re-recorded imitation and not a different model wearing the name. The Kokoro credit and thanks are in the app's Settings screen.
Does it download the model on first launch?
No. The weights ship inside the app bundle, which is why the download is large. After installing, Sandbook never fetches the model, a voice, or a phoneme table. Reading, speech, scanning and export all work offline; only adding a web page needs a connection.
Can I use Kokoro's non-English voices?
Yes, for Spanish, French, Italian, Brazilian Portuguese and Hindi. Sandbook bundles 41 Kokoro voices: 28 in US and UK English plus 3 Spanish, 1 French, 2 Italian, 3 Brazilian Portuguese and 4 Hindi. The voice decides the language: a book in one of those languages opens with a matching voice, and each book remembers the voice you pick. Voices in other languages are not included.
How does on-device Kokoro compare to a server for speed?
Synthesis is sentence-by-sentence with a look-ahead buffer, so audio starts after the first sentence rather than after the whole document. In practice a paragraph starts speaking almost immediately and generation stays ahead of playback. A longer document doesn't make it start later, because there is no queue to wait in.
How long until it starts speaking?
On an iPhone 15, about 2.5 seconds after you press play on a fresh chapter, measured in October 2026 with the earlier MLX engine, before the iPhone app moved to Core ML; no figure has been published yet for the Core ML engine. Sandbook generates the first sentence, starts playing it and renders the rest behind it, so the wait does not grow with the length of the chapter. Older supported iPhones, back to the iPhone XS and XR, take longer.
What hardware does it need?
iOS 18.2 or later on an iPhone XS, iPhone XR or later, or an iPad with an A12 chip or later; or macOS 15.2 or later on a Mac with Apple silicon (M1 or later). Every iPhone and iPad that can install it runs the voices; Intel Macs are not supported. It does not run on iOS 17.
Is there a Kokoro TTS app for iOS?
Yes: Sandbook, free on the App Store for iPhone and iPad. It bundles Kokoro-82M and all 41 voices, runs them on the device and works offline from the first launch, airplane mode included. What you read never leaves your device. It needs iOS 18.2 on an iPhone XS, iPhone XR or later and does not run on iOS 17. The same app runs natively on Macs with Apple silicon (macOS 15.2 or later), under the same free listing.
Is there a Kokoro TTS app for Mac?
Yes: Sandbook for Mac, free under the same App Store listing as the iPhone app. It bundles Kokoro-82M and all 41 voices, runs them on the Mac with Core ML, the same engine as on iPhone and iPad, and works offline. It needs a Mac with Apple silicon (M1 or later) and macOS 15.2 or later; Intel Macs are not supported.
Sources
- Kokoro-82M model card (hexgrad), with its voice list — Hugging Face · checked November 11, 2026
- Apache License 2.0, the licence Kokoro-82M is published under — apache.org · checked November 11, 2026
Kokoro TTS, on your own device
A Kokoro TTS app that keeps the whole model — and everything you read with it — on your own device. 41 voices in six languages, on iPhone, iPad and Mac, free with nothing to buy inside.
iOS 18.2 or later · iPhone XS / XR or later · iPad with A12 or later · Mac with M1 or later, macOS 15.2