Ollama brings text-to-speech models to Apple Silicon Macs
The update introduces Gradium’s Phonon models, with five languages supported and Windows and Linux availability to follow.
Mac users can now generate speech locally with Ollama using Gradium’s Phonon models, the platform announced. The capability requires an Apple Silicon Mac and version 0.40.11 or later of Ollama. Support for speech generation on Windows and Linux is coming soon.
This is the platform’s first speech-generation offering. Ollama says listeners can hear the audio while it is being generated, without waiting for a complete response. Phonon is available to download within Ollama.
Users have a choice of separate models for English, Portuguese, French, Spanish and German, or one multilingual model covering those five languages. Available voices depend on the model, with options including British and American English accents.
Ollama reported speech generation that was 35% faster than Gradium’s implementation in testing using a MacBook Pro equipped with M5 Max. It attributed the performance to Apple’s MLX framework and said its word error rate matched that of the Hugging Face Transformers reference implementation.
The open-source platform also runs compatible AI models on a user’s own computer, allowing offline use without sending data to external servers. A Mac’s hardware affects both performance and which model sizes it can handle.
