The Hugging Face model card describes phonon-31m-en as a 31-million-parameter English model with 64.5 MB of 8-bit quantized weights. It generates mono 24 kHz audio and supports streaming output, with stated deployment targets including browser tabs, phones and laptop CPUs.
The family also lists 40M models for English, French, German, Spanish and Portuguese, plus a multilingual model covering those five languages. The model card lists Apache 2.0 for phonon-31m-en and CC BY-NC 4.0 for the listed 40M variants. Gradium reports that its models lead its published Elo and word-error-rate comparisons in each language; those results are the vendor’s own evaluation claims, not independent validation. The card points to runtime options for browser, mobile, server and other integrations.