macOS text to speech for AI agents: use your Mac's built-in voices

What are macOS voices?

They are the system voices macOS uses for its own spoken content, such as reading selected text aloud. The /usr/bin/say command speaks with any installed voice and can save the result to an audio file instead of playing it, which is what Ork does.

Voices come in three tiers: Compact, which is small and robotic; Enhanced; and Premium, the most natural, available for a handful of voices such as Ava and Zoe in US English and Jamie and Serena in British English. Siri's voices aren't available to say. Personal Voice, the voice you can record of yourself, is only available to apps you grant permission to.

How to use text to speech on a Mac

In macOS 26 and later, Apple renamed the Spoken Content settings to Read & Speak. That's where you pick the system voice and download better ones; older guides still use the old name.

1. Open System Settings → Accessibility → Read & Speak. On macOS 15 and earlier, it's Accessibility → Spoken Content.
2. Choose a voice from the System voice menu. To get Enhanced or Premium voices, click the Info button next to it, pick a language, and click a voice to download it.
3. Turn on Speak selection to have the Mac read any selected text with a keyboard shortcut, Option-Esc by default.
4. In Terminal, say -v '?' lists installed voices, say -v 'Zoe (Premium)' "hello" speaks, and say -o hello.aiff "hello" saves the audio to a file.

What it costs

Nothing. The voices come with macOS, run on the Mac's own processor, and have no usage limit. That's the whole case for them: a bot that reads replies all day costs the same as one that never speaks. The trade-off is quality, since even Premium voices sound less natural than the best hosted ones, and there's no transcription.

Prices in US dollars from each provider's pricing page, checked November 2026. An hour of speech is taken as about 54,000 characters (150 words a minute); real voices run a little faster or slower. Fish Audio bills per UTF-8 byte, which is one byte per character in English.

What you can do with macOS voices in Ork

The server renders each reply into a short audio clip with say and sends it to whoever is listening, so it plays on your phone or laptop like any other voice, not out of the server's speakers. The voice and calls guide covers each feature.

• Hear replies. Choose Speak this reply on a message, or turn on Read replies aloud in a bot's settings.
• Call a bot with Call at the top of the chat, or have a bot call you when you ask it to.
• Talk instead of typing, using on-device speech recognition. macOS voices don't transcribe, so your words are recognized by your browser, such as Chrome, or by the Ork desktop app on a Mac.
• Choose any installed voice, and a different one for each bot. Premium and Enhanced voices are listed first.

When to choose macOS voices

Choose them when you run Ork on your own Mac and want bots to speak right away, with nothing to sign up for, nothing to install and nothing to pay. Speech is made on the machine running Ork, with no vendor involved, then sent to whichever device is listening.

For voices that also transcribe, a self-hosted speech server keeps everything on your own hardware and works on Linux too; on a Mac, mlx-audio runs Kokoro and Whisper quickly. Hosted providers sound more natural and need a key: ElevenLabs and Fish Audio clone voices, OpenAI is the simplest, and Cartesia is built for low latency. The voice provider comparison shows all six.

Limits

• Ork only offers Server (Mac) when its server process runs directly on macOS. A server in Docker, including Docker Desktop on a Mac and the desktop app's This computer install, runs Linux inside the container, so the option doesn't appear. Linux hosts don't list it either.
• It doesn't transcribe. Hold-to-talk and calls depend on speech recognition in your browser or the desktop app, and voice notes from messaging apps reach bots without a transcript.
• macOS voices sound less natural than the best hosted voices, and there's no voice cloning.
• The voices are whatever is installed on the server's Mac. A newly downloaded voice can take a few minutes to show up.
• Ork lists the voices say -v '?' reports. We haven't tested Personal Voice, which needs the server process to be granted permission.
• On the mobile app, calls can use your phone's own voices if you pick a Device voice in the bot's settings. Otherwise replies use this provider.

How to set it up

This is for someone running Ork from source on their own Mac. The self-hosting guide has the full instructions.

1. On the Mac, clone Ork, copy .env.example to .env, then run pnpm install, pnpm sandbox:build and pnpm dev. This runs the Ork API directly on macOS.
2. To use hold-to-talk and calls, set PHONE_CALLS_ENABLED=true in .env and restart.
3. Optionally download Enhanced or Premium voices under System Settings → Accessibility → Read & Speak, with the Info button next to System voice.
4. In Ork, open Settings → Voice and set Provider to Server (Mac). There's no key to enter.
5. Pick a Voice and choose Hear a sample.