Ollama Mac App Adds Offline Text-to-Speech With MLX
Ollama's Mac application has updated to support offline text-to-speech generation using Gradium's Phonon model powered by Apple's MLX framework. This new feature delivers up to 35% faster performance on Apple Silicon compared to standard reference implementations while keeping all data processing local. This update expands local AI capabilities beyond reading and writing into speech generation, providing an offline solution that ensures user data privacy. Developers and privacy-conscious users can now implement responsive, real-time voice synthesis directly on Apple hardware without relying on cloud-based services. The integration supports the Phonon model family across five languages and various accents, enabling real-time audio generation as text is being produced. While currently available on macOS via Apple's unified memory architecture, Windows and Linux support will follow in subsequent updates.
## BACKGROUND
Ollama is an open-source platform designed to easily run and manage large language models locally on consumer hardware. Apple's MLX is an open-source array framework optimized for machine learning on Apple Silicon, utilizing the Metal framework directly for high-performance GPU computing without translation layers.