Written by Pratik Nayak, the developer of Readlet. This guide covers free options as well as our app. Updated November 6, 2026.
What Kokoro is
Kokoro-82M is a text-to-speech model with 82 million parameters, released with open weights by hexgrad. It runs comfortably on a modern Mac without a cloud service, and its voices sound far more natural than older system voices.
Run it yourself
If you are comfortable with Python, projects such as kokoro-onnx let you install the model and generate speech from scripts. You manage the Python environment, the model download and how text reaches it. It is a good route for developers and automation.
Readlet: Kokoro in a menu bar app
Readlet is a Mac app that uses Kokoro to read what you choose, without Python or Terminal.
- Full-quality voice. Readlet uses the full-precision Kokoro v1.0 model, with the Heart, Bella, Emma and Adam voices.
- Simple setup. The first launch asks before downloading the voice package, about 392 MB, with progress, cancel and retry.
- Starts quickly. Readlet speaks the first sentence while it prepares the rest, so you are not waiting for a whole document.
- Reads what you point at. Drag over text on screen, press Control–Option–R for selected text, or open a text file, RTF, DOCX, searchable PDF or image.
- On your Mac. After setup, speech and text recognition run locally. Your text is not uploaded.
The app and speech worker source are included under GPL-3.0-or-later.
Limits worth knowing
- Readlet's voices are English.
- It requires an Apple Silicon Mac with macOS 26 or newer.
- Complex PDF layouts, tables, tiny text and low-quality scans may not read correctly.
- Control–Option–R is intended for selected-text reading, but app compatibility still needs broader testing.