palshub/phonon-31m-en-coreml

Phonon 31m English: precompiled Core ML bundle

A mirror of Gradium's Phonon 31m Core ML bundle, with every .mlpackage precompiled to .mlmodelc, so iOS apps can download it file by file and load it without compiling on the device. It is used by @pocketpalai/react-native-speech-phonon (iOS 18+, Apple Neural Engine).

Contents

• phonon-31m-en-coreml/: the bundle (bundle.json, the compiled flow and Mimi models, host.safetensors, tokenizer.json, and 38 English voices in voices/).
• phonon-31m-en-coreml.manifest.tsv: path<TAB>size<TAB>sha256 for every file in the bundle. Apps verify each download against it.

Provenance

• Source: the phonon-31m-en-coreml.zip asset of the upstream gradium-ai/xn-ptts v0.4.0 release, sha256-checked against the pin in phonon.lock.json.
• Model: hf://gradium/phonon-31m-en@8acf8881a96315724c72ed3aad2a0900d1b230ba. Voices: hf://gradium/phonon-voices@da6c9ec0d47be8df5c87c730a3aa97187f3229c3 (see phonon-31m-en-coreml/source.json).
• Precompiled with xcrun coremlcompiler by the Phonon binaries workflow, using scripts/precompile-coreml.sh. The weights are not modified.

License and attribution

Phonon 31m and its voices are released by Gradium under the Apache License 2.0. Phonon builds on Kyutai's Pocket TTS and the Mimi codec. This mirror redistributes them unchanged apart from the Core ML compilation. The larger Phonon 40m models are CC BY-NC 4.0 and are not included here.