Ollama adds offline AI voice generation for mac, boosting speed by 35%

Ollama, an open-source AI platform, has introduced offline text-to-speech (TTS) capabilities for its Mac application, with Windows and Linux versions to follow. The feature allows users to generate AI voices locally without relying on cloud services, enhancing privacy and performance. The Phonon series by Gradium is the first supported voice model, offering smaller size and advanced technical specifications. On macOS, Ollama utilizes Apple's MLX framework to optimize performance on Apple Silicon. Testing on a MacBook Pro with an M5 Max chip showed Phonon's speech generation speed was 35% faster than Gradium's own implementation while maintaining comparable word error rates. The new feature supports real-time voice generation, allowing users to hear speech as it is being generated, improving the conversational experience. Phonon currently supports five languages and offers various accents, including American and British English. Ollama, with over 185,000 stars on GitHub, enables users to run open AI models locally or access cloud-based models as needed.