Ollama for Mac can now turn text into speech on-device

What is new

Ollama, the app for running AI models locally, now supports text-to-speech on the Mac, with Windows and Linux support described as coming soon. 9to5Mac reports that this is the first time the platform has supported speech generation ^[1].

The first speech model is Phonon, a family of small text-to-speech models from Gradium. Ollama describes it as state of the art ^[1]. Speech is generated in real time, so users can listen to AI-generated responses as they are produced instead of waiting for the full reply ^[1].

How it works

Ollama lets people download open AI models and run them on their own Macs, or reach more powerful models through the cloud. On macOS it uses Apple's MLX, an open-source machine learning framework that takes advantage of Apple Silicon's unified memory so supported models run faster and more efficiently ^[1].

According to Ollama, MLX delivered 35% faster speech generation than Gradium's own implementation in tests on a MacBook Pro with an M5 Max, while matching the word error rate of the reference implementation in Hugging Face Transformers ^[1]. These are the company's own test results, and the figures come from one machine.

Phonon covers five languages: English, French, German, Spanish and Portuguese. There are separate models for each language, plus a multilingual model that handles all five. Users can pick from voices with different accents, including American and British English, depending on the model ^[1].

Availability

The speech models are available now for Apple Silicon Macs running Ollama 0.40.11 or later. Users download Phonon through Ollama, and there are instructions and code examples for using the models through its API ^[1].

Why it matters

For businesses, the appeal is privacy and control. Running models locally means people can use AI without an internet connection or sending data to external servers, although 9to5Mac notes that performance and the size of model you can run depend heavily on your Mac's hardware ^[1]. A company that wants to read documents aloud, create audio versions of internal updates or add a voice to a prototype can now do so without paying per character to a cloud service or sharing text with a third party.

The real-time behaviour is also useful. Hearing an answer as it is written feels closer to a conversation, and it can make accessibility features and hands-free tools more practical.

What to keep in mind

Small local models will not suit every job. Quality, speed and voice choice will vary by Mac, and the language list is limited to five. Teams with strict requirements should test the voices on their own content and hardware before relying on them.

Ollama itself is open source and has become one of the most popular AI projects on GitHub, with nearly 185,000 stars and around 18,600 forks, according to 9to5Mac ^[1]. That popularity suggests many developers will start building on the new speech support quickly.

The change is a reminder that useful AI is increasingly something that runs on the laptop in front of you. Text and speech are now both available locally in the same tool, which lowers the barrier for small teams experimenting with voice features.

Source: 9to5Mac — Ollama for Mac gains support for local text-to-speech models