Student
Study from notes, textbooks and papers.
- Classroom pacing
- Spoken maths
- Slow down or repeat an equation
Preset classroom (coming)
Study narrator · open source · self-hosted
Listen to the papers you have to read.
Drop in a PDF (a paper, a textbook chapter, lecture notes), pick a voice and a speed, and listen with the text highlighted as it plays. Formulas are spoken as a person would read them aloud rather than skipped, and technical vocabulary is pronounced correctly.
The value x in the given equation comes from a normal distribution with mean μ and standard deviation σ. If X is a normally distributed random variable and X ~ N(μ, σ), then the z-score is:
z =x − μσ
Spoken as“z equals the fraction x minus mu over sigma, end fraction”
Who it's for
Made for three ways of studying
One engine, three ways of listening. They differ in pacing and in which parts of the page are read; one-switch presets for each are on the way.
Technical learner
Learn from technical books and papers.
- Code listings cued, acronyms read well
- Equations
- 1.5× to 2× listening
Preset study (coming)
Book listener
Narration while you follow the printed book.
- Natural pacing, no extra pauses
- Chapter navigation and resume
- Captions and footnotes muted
Preset listen (coming)
How it works
From the page to your ears, one sentence at a time
The hard part is the reading, not the voice: the page is taken apart into typed blocks, put back in order, and turned into words a voice can say.
Features
Built for the pages that other readers mangle
Papers and textbooks have columns, footnotes, symbols and jargon. YomiKoe is built around them.
Reading order that holds
Two-column pages read in order; headers, footers and page numbers are left out.
Spoken maths
Display and inline equations read as English: fractions, powers, roots, sums, integrals, Greek letters.
Terms said right
An academic pronunciation lexicon, plus “pronounce as” on any word you select.
Word-level highlight
The current word and sentence are highlighted in step with the audio.
Ten voices, 0.5× to 3×
American and British English, female and male; speed from 0.5× to 3× in 0.05 steps.
Repeat an equation
Skip or repeat the current equation with one action; equations play at 0.8× by default.
A library that resumes
Your documents, reading position, voice and speed are kept; come back tomorrow and carry on.
Keyboard and lock screen
Space, arrows and +/− on the keyboard; lock-screen controls on phone browsers.
Mute what you skip
Turn off footnotes, captions or the references section with a toggle.
Open source
Open, and yours
Commercial readers do this well, but they are subscription-gated, closed, and run on their own voices. YomiKoe is an open reading engine: the hard parts, text extraction and playback, are in the open, and the voice is a pluggable backend.
The voice is a backend you can swap; the reader never changes.
| Topic | YomiKoe | Subscription readers |
|---|---|---|
| Cost | Free and open source | A subscription |
| Where it runs | On your own machine, CPU only, via Docker Compose | On the vendor's service |
| Voices | Open models behind a swappable backend; Kokoro first | Their own, closed voices |
| Your documents | Stay on your machine, in a local library | Held by the service |
Getting started
Run it on your laptop
One image, one service. You need a container runtime with Docker Compose v2.
- Clone the repository.
- Build and start with Docker Compose. The first run downloads the Kokoro-82M voice model, which takes a minute or two; later starts run offline.
- Open
http://localhost:8765/and drop in a PDF.
# 1. clone
git clone https://github.com/kris-jskk/yomikoe.git
cd yomikoe
# 2. build and start (Ctrl-C stops it)
docker compose up --build
# 3. open http://localhost:8765/ in your browser
Questions
Good to know
Do I need a GPU?
No. YomiKoe is built to run locally on a laptop with no GPU: the first voice backend, Kokoro-82M, runs on the CPU. A GPU voice is planned as an optional second backend.
Which PDFs does it read?
Born-digital and scanned PDFs. YomiKoe checks each page for a text layer and runs OCR only on the pages that have none.
Is there a hosted version?
Coming. A hosted reader at app.yomikoe.app is part of the hosting milestone. Until then you run YomiKoe yourself with Docker Compose.
What is the licence?
YomiKoe is open source, built on a permissive stack: every dependency and model is under Apache-2.0, MIT or BSD terms, with no copyleft code shipped. YomiKoe itself is released under the Apache License 2.0.