Why it exists
A browser tab is the wrong place to convert a twelve-hour novel: there are
no on-disk checkpoints, so closing the tab or a crash halfway through
loses the work, and WebGPU in a tab is fragile. The native tools
aren’t a fit either: Apple Books’ “read aloud” only
plays live and can’t export files, and macOS’s built-in
say sounds robotic next to a neural voice.
aloud is the only one that gives you Kokoro’s specific voice, exported to files, resumable, GPU-accelerated, and fully offline: a real audiobook factory rather than a live reader. It takes the part of reading a book aloud that most wants to “set it and forget it” and puts it in the one place that can do that well: a desktop app with real disk and real background work.
What it does
- Drop an
.epub,.pdf, or.txt; it finds the chapters (EPUB spine / table-of-contents logic) and lists them as a table of contents you can check and uncheck. Convert the whole book or just the chapters you want. - Converts each selected chapter to AAC
.m4aon-device, then stitches them into one.m4baudiobook with embedded chapter markers and the title and author: the file Apple Books wants, with navigable chapters and resume-where-you-left-off. - Writes every file straight into a folder you pick. Nothing is uploaded; after the model downloads once it works entirely offline.
- Survives a quit or a crash: a per-book checkpoint on disk means relaunching offers Resume, and it re-renders only the chapters that weren’t finished, never redoing work.
A neural voice, on the GPU
The voice is Kokoro-82M, an 82-million-parameter neural
model that runs entirely on your machine. The difference from the native
tools is where it runs. An earlier attempt built this as a native Swift
app, but that path structurally lost the GPU: ONNX Runtime’s CoreML
compiler couldn’t build this Kokoro graph and could only ever fall
back to the CPU.
Electron bundles its own Chromium, so the WebGPU path lights up on any Mac. The GPU CoreML couldn’t give us is the one this app uses by default. And it’s no longer a guess: a badge in the corner of the window shows the actual compute backend (GPU · WebGPU with the adapter name, or CPU · WASM with the reason it fell back), so you can always see whether it’s on the GPU.
Opening it on macOS
aloud is a free, open download, signed only ad‑hoc rather than notarized with a paid Apple Developer account. So the first time you open it, macOS shows its usual “unidentified developer” prompt: right‑click the app and choose Open (then Open in the dialog), and it remembers your choice from then on. That one‑time step is the only thing standing between the download and your first audiobook.