Supporting case
Ollama adds local speech, with a commercially usable English model and narrower multilingual licences
What happened
Ollama's October 24 release adds text-to-speech through Gradium's Phonon family. This can read an assistant's text aloud on an Apple Silicon Mac instead of sending each speech request to a hosted provider. The initial Ollama route is macOS-only; Windows and Linux support are described as forthcoming.
The small English checkpoint has 31 million parameters excluding its audio tokenizer and uses Apache 2.0. The 40-million-parameter models, including the multilingual version, use a noncommercial licence. Downloadability therefore does not mean equal commercial freedom in every language. Model registry ↗; Gradium model card ↗. Reported memory and speed are developer measurements on specified hardware, not independent tests on ordinary devices.
Independently running speech output can remove one cloud dependency from an assistant. Hardware, operating-system support and licence restrictions limit who can use that capability commercially, especially across languages.
Dates and implementation stage
Development 2026-10-24
