Gradium releases weights for Phonon on-device text-to-speech

Gradium released 31M- and 40M-parameter Phonon models. The English 31M model uses Apache 2.0, and the 40M language-specific models and five-language model use CC BY-NC. Phonon generates 24 kHz audio. Gradium measured about 7 ms to first audio for the 31M model on a MacBook Air M5 CPU.

Visit original source