ChorusReader
Every character gets a voice
Takes your books and gives every character a distinct voice, entirely on-device. Three engines: Standard, with 40+ voices in seven languages, free; Personal, which clones any voice from 15 seconds of audio; and Premium, the most expressive, with multi-voice character dialogue. Version 3.2 lights up each word as it is read, added Mandarin, and moved five languages onto Chorus's own pronunciation engine. A bookstore of 100+ hand-picked free classics is built in.
Why multi-voice
Audiobook narration is a solved problem if you're fine with one voice reading everything. But characters have voices. ChorusReader works out who says each line and gives every character their own voice, while the narrator keeps theirs. All of it runs on your iPhone or iPad: no account, no upload, no per-character fees.
Capabilities
- Three engines — Standard: 40+ voices in English, Spanish, French, Italian, Portuguese, Hindi and Mandarin, free. Personal: clone any voice from 15 seconds of audio. Premium: the most expressive narration, with multi-voice character dialogue
- Fully Voiced — the cast you pick is the cast you hear; each line stays in its speaker's voice, including quotes that run across paragraphs and the « » and dialogue dashes of French and Spanish books
- Words light up as they're read — every word highlighted in step with the narration, a character at a time in Chinese
- Starts in seconds — audio streams while it generates, and the rest of the book is narrated while you listen
- Built-in bookstore — 100+ hand-picked free classics with a sample narrated on the spot, a curated list of open-access research papers with the reason each was chosen, and your RSS feeds
- Reads what you have — EPUB, PDF, plain text, Markdown, Word and Pages files, photos of pages, and any article shared from Safari
- Private by design — no account, and nothing you read, record or generate leaves the device
Demo
From page to voice
A book goes through three stages on the device. Analysis finds the chapters, the quotes and who is speaking. One of three engines turns each passage into audio. Playback starts as soon as the first passage exists.
book analysis: Chapters (from EPUB, PDF, text), then Quotes (“ ”, « », dialogue dashes), then Cast (who says each line). three engines, all on-device: Standard (Kokoro-82M · phonemes from ChorusG2P), Personal (Pocket TTS · a voice cloned from 15 s), Premium (Chatterbox Turbo · multi-voice dialogue). playback: Streaming (starts in seconds, narrates ahead), then Word highlight (each word lit as it is read)
- ChorusG2P is Chorus's own pronunciation engine: Spanish, French, Italian, Portuguese, Hindi and Mandarin. English runs on Misaki.
Key decisions
-
Our own pronunciation engine
six languages of text-to-phoneme rules in Swift, scored against the engine it replaced as a black box, on held-out prose by a different author
-
Listening tests, not just word matches
where the two engines disagree, both are rendered through the same voice and the audio is measured; that is how French nouns in -ent got their vowel back
-
Cut a passage by what it costs the voice
a 326-character Hindi paragraph came to 743 phoneme tokens against a limit of 510, so long passages are split at the cheapest boundary by token count, not by length
-
Voice licences are a build phase
every voice file is listed by hash with its source and licence, and the build fails on anything else; it also found a model nobody loaded, and the app got about 300 MB smaller
-
One crash is a class
a trap on one word becomes a fixture of every word like it, a scan of the source for the pattern, and a fuzz run on the app's real stack size
-
No account, no server
narration happens on the device, so there is nothing of yours to see
-
One purchase, never a subscription
the free tier is the whole Standard engine and the entire library