Text to speech
Works in any app. A natural voice, made on your Mac, reads it with each word marked as it’s spoken.
Sample textRyan
The lighthouse keeper climbed the spiral stairs each evening, counting every step out of habit. At the top, he wiped the salt from the glass and lit the lamp.
Choose a voice to hear this read aloud
What it does
OrbOS reads the text you select, in any app, aloud in a natural voice generated on your Mac. The notch, a small panel at the top of the screen, shows what it's reading. In apps that report where their words are, the spoken word is highlighted right in the app, scrolled into view as it goes.
The shortcut is yours to change, and Read Aloud is in the right-click Services menu too.
Features
Anywhere you can select text. OrbOS takes the selection through macOS accessibility, or copies it when an app won't say, and puts your clipboard back.
In Safari, Mail, Notes, TextEdit, Preview, Pages and Chromium-based apps. Five styles, in the voice's colors or one you pick.
Ryan, Aiden, Nora, Eleanor, Maya, Theo and Serena, with styles such as calm, enthusiastic or news. The orb takes the colors of the voice that's reading.
A pill-shaped panel at the top of the screen, with play, pause, speed and stop. It opens to show the text, and never takes focus from your app.
OrbOS scrolls the app to keep the spoken word in view, and waits while you scroll yourself.
Record a passage or choose a file. OrbOS picks the best 5 to 15 seconds, checks it, transcribes it on your Mac and clones the voice.
Tell it how to say a word. The highlight stays on the word as written. Import and export your list as JSON.
Readings are kept with their text and audio, by day, with search. Replay one with the spoken word highlighted, or save it as an .m4a. 30 days by default, or as long as you like, or not at all.
Speed from 0.75× to 2× without changing the voice's pitch, and skips back and forward. The keys are held only while a reading plays or is paused, so other apps keep them the rest of the time.
Voices
Seven voices come with OrbOS, each with one or more styles. Each sample is the voice reading the same lines, made on a Mac by OrbOS's speech engine.
Ryan
Cobalt · Neutral, Enthusiastic, Calm, News
Cobalt
Brisk, dynamic male voice with strong rhythmic drive.
Sunlit
Sunny, friendly American male voice with a clear midrange.
Rosewood
Warm, smooth American female narrator.
Lagoon
Poised British female voice with crisp diction.
Fuchsia
Bright, upbeat young female voice.
Fern
Soft-spoken, thoughtful British male voice.
Amethyst
Warm, gentle young female voice; Chinese is its native language.
No cloud voice, no server in between. OrbOS reads with models that live on your Mac, so what you read is yours alone.
The voice is made on your Mac's GPU, inside OrbOS. Nothing you read is sent anywhere.
Nothing to sign up for, and nothing reported back about how you use it.
Once its models are on your Mac, reading needs no network at all.
To download a model when you ask, for speech recognition the first time you add a voice in a new language, and for a daily check for a new version of OrbOS, on by default, which you can switch off.
Kept on your Mac for 30 days by default, for a day, a week, forever, or not at all.
Only how long a reading was and which app it came from.
OrbOS's speech engine is written in Swift on MLX, Apple's machine learning framework, and runs the voice on your Mac's GPU, inside the app. Nothing else to install: you download the models in the app.
Measured on a MacBook Pro with M5 Pro (20-core GPU) and 64 GB of memory, macOS 27. First word and speed: OrbOS's built-in engine reading three sample texts (86 to 1,306 characters) at 1×, medians of 15 readings, models already loaded. By model: eight Qwen3-TTS variants, generation alone. Your Mac's numbers will differ.
Read the code, build it yourself, report a problem or send a change. Everything that makes OrbOS is in one repository.
FAQ
No. The voice is made on your Mac, and nothing about your text leaves it. There is no account and no telemetry.
Yes, once its models are on your Mac.
It reads selected text in any app: through accessibility, the Services menu, or a copy of the selection. Highlighting the spoken word in the app itself needs the app to report where its words are, as Safari, Mail, Notes, TextEdit, Preview, Pages and Chromium-based apps do. Elsewhere, the notch still shows the words.
English, Chinese, German, Spanish, French, Italian, Portuguese, Russian, Japanese and Korean. OrbOS tells the language from the text, or you can set one.
Up to 10,000 characters at a time, about 10 minutes of speech.
Yes. Speed is applied as it plays and keeps the voice's pitch, from 0.75× to 2×.
Yes: record a passage or choose an audio file, yours or that of someone who has agreed to it. OrbOS picks the best 5 to 15 seconds, transcribes it on your Mac and clones the voice from it. Your voices stay on your Mac.
The voice runs on the Mac's GPU with MLX, Apple's machine learning framework for Apple silicon.
The app itself uses about 25 MB when idle. The speech engine uses about 6 to 9 GB while reading, less with smaller models, and can free it between readings.
macOS 26 or later, on Apple silicon