Speech-to-text apps have flooded the App Store, making it easier than ever to type on our Macs with just our voice. I have previously covered one such app called TypeWhisper. However, there has been a dearth of good text-to-speech models, and most of the good ones are locked behind a subscription.
Ollama, an open-source app that lets you run local AI models on your computer, wants to change that. It works on macOS, Windows, and Linux, and your prompts are never stored or used for training.
How is Ollama changing that?
Ollama now supports text-to-speech models on the Mac, starting with Phonon. Built by Gradium, Phonon is a family of small models that can turn any text into speech right on your Mac. Windows and Linux support is coming soon.
The voices featured on Ollama’s blog sound really natural. You can choose from voices like Zoey, Easton, Freya, and Hugh with American and British accents. Apart from English, there are models for French, German, Spanish, and Portuguese.
Speed shouldn’t be an issue either. Thanks to Apple’s MLX framework, Ollama claims Phonon runs 35% faster than Gradium’s own runtime on a MacBook Pro with M5 Max, without losing accuracy.
How can you use it?
You can pair Phonon with a chat model to have its replies read aloud. Playback begins after the first sentence, so you don’t have to wait for the full answer.
To get started, you need an Apple Silicon Mac running Ollama 0.40.11 or later. Just pull the Phonon model, and you are good to go. Note that while the English model is free to use under Apache 2.0, the French, German, Spanish, Portuguese, and multilingual models are limited to noncommercial use.
RELATED COVERAGE:
