MacEpic/phonon-31m-en-coreml

Phonon 31M English, Core ML

Gradium's Phonon 31M English text-to-speech model, exported to Core ML for Apple silicon, with its 37 English voices from gradium/phonon-voices. MacEpic hosts this copy unchanged so TextSniper can download it at a fixed revision; nothing here is MacEpic's own work.

Contents

The unzipped phonon-31m-en-coreml.zip of gradium-ai/xn-ptts v0.6.0 (the release asset's SHA-256 is 2869326ab074f55ed5b644e68b1befef69dc51545b73296d49f5b4dab14c4218), as its exporter wrote it:

• bundle.json: every file with its size and SHA-256, and the model's dimensions.
• flow-c414-t1.mlpackage, flow-c414-t48.mlpackage, mimi-c266.mlpackage: the Core ML programs.
• host.safetensors, tokenizer.json: the host-side tensors and the tokenizer.
• voices/: the English voices.
• source.json: the exact sources, gradium/phonon-31m-en@8c64b55 and gradium/phonon-voices@da6c9ec.

Use

Load it with xn-ptts' Swift package (ptts, module PhononTTS, v0.6.0, macOS 15 or iOS 18 on Apple silicon):

let models = try await PhononModels.download(
    from: URL(string: "https://huggingface.co/MacEpic/phonon-31m-en-coreml/resolve/<revision>/")!
)
let phonon = try await Phonon.load(models: models, language: .english)

download(from:) reads bundle.json, fetches each file and checks its size and SHA-256.

License and credit

Apache License 2.0 (LICENSE), as the model and the voices are published by Gradium. The model is by Gradium; its runtime, xn-ptts, builds on Kyutai's Pocket TTS.