Ollama introduces local text-to-speech on Macs with Apple Silicon
Ollama now supports local text-to-speech using Phonon models on macOS with Apple Silicon. It offers streaming playback and an API compatible with the OpenAI API. Licensing varies by model variant.
Ollama has introduced local text-to-speech using the Phonon family of models from Gradium. Running it requires a Mac with Apple Silicon and Ollama version 0.40.11 or later. Support for Windows and Linux is planned; the source does not specify a date.
The English model phonon is licensed under Apache 2.0. The 40M variants are licensed under CC BY-NC 4.0, which restricts use to non-commercial purposes. Separate variants support French, German, Spanish and Portuguese; the phonon:40m-multilingual model includes all five languages, including English. Voices can be selected from the Gradium catalog.
The integration provides the /v1/audio/speech endpoint, which is compatible with the OpenAI API. According to Ollama, existing code using the SDK from OpenAI only needs its base URL changed. Audio is streamed as it is generated; when a response from a chat model is passed in, playback can begin after the first sentence. The output is mono audio at 24 kHz, in WAV format by default, or optionally as PCM with 16-bit samples.
On Apple Silicon, Ollama uses acceleration through MLX. According to Ollama, it was 35 % faster on a MacBook Pro with M5 Max than the runtime from Gradium and achieved the same word error rate as the reference implementation in the Hugging Face Transformers library. These measurements were published by the provider itself.
Why it matters
The new support allows text and responses from chat models to be converted to speech as they are generated on a supported Mac. It gives developers a local interface for speech output that they can connect to code using the SDK from OpenAI. For commercial deployment, the difference between the license for the English model phonon and the non-commercial license for the 40M variants is crucial.
Two audiences, two different impacts
What this means
For individuals
Users with a supported Mac can have text or responses from a chat model read aloud locally and save the generated speech to a WAV file.
For a business
Development teams can add local speech output through a compatible API. When choosing a language variant, they must take into account that the license for the 40M variants restricts use to non-commercial purposes.
DevelopmentCheck the original
Event sources
clearly official source · 1 publisher, 0 independent. We count feeds from the same owner only once.