It reads to you, in a voice made on your Mac.
Select any text, in any app, and press fn and space. Your Mac reads it aloud in a natural voice generated on your own machine, and the words light up as they are spoken. No cloud voice, no robot monotone.
abstract
Tesseract gives your Mac a voice worth listening to. An article, a contract, a long reply you want to hear before you send it: select it, press the keys, and listen, following along in the overlay or walking away. The voice itself is not picked from a system menu. It is generated, and you decide what it sounds like by describing it in words. The same voice can speak the assistant's answers out loud. Everything is synthesized locally, so what it reads to you stays between you and the machine.
keywords: select, press, listen · a voice you design · no cloud voice
theorem 3.1 (read aloud)
Select, press, listen.
Proof. Any text in any app can be read to you: select it and press fn and space. A quiet overlay appears and the words light up one by one as the voice reaches them, so you can follow along, check a detail, or leave it playing while you do something else.
in plain terms
- ·fn space on any selection
- ·words light up as read
- ·works in every app
theorem 3.2 (a designed voice)
Describe a voice, and it exists.
Proof. The voice engine is a voice-design model. Instead of choosing between a handful of presets, you write a sentence about the voice you want, warm and unhurried, quick and dry, and that is the voice your Mac speaks with. It is generated on your machine, in the language you choose.
in plain terms
- ·a voice from a written description
- ·generated locally
- ·~4.2 GB model
theorem 3.3 (the long read)
Steady through the long read, and it starts without waiting.
Proof. Synthetic voices tend to drift and wobble as a passage grows. This one stays steady through long answers and whole articles. And playback begins while the rest is still being generated, so a long read starts like a short one.
in plain terms
- ·long passages hold steady
- ·playback starts early
- ·speaks chat answers too
fig. 01: the read. The highlight walks the sentence as the voice speaks it, and generation runs ahead of playback.
the hotkey
press fn and space with text selected · changeable in Settings
the voice
describe it in a sentence in Settings, or keep the standard one
the model
a voice-design model, ~4.2 GB, generated entirely on your Mac
the assistant
one switch makes every chat answer spoken aloud
§ endsee for yourself
The voice is shipped and in daily use. It runs on any Apple silicon Mac with 16 GB of memory, and what it reads to you is never sent anywhere.
next in the series: paper no. 04 · chat →