Kokoro-82M Core ML (ANE, background-safe)

Kokoro-82M converted to Core ML with laishere/kokoro-coreml (fp16 + int8 palettization), packaged for the KokoroANE Swift pipeline.

  • coreml/ — the seven pre-compiled .mlmodelc stage bundles (KokoroAlbertKokoroTail). No Core ML compilation is needed on device; download and load.
  • voices/ — the 28 English voice style tables as raw little-endian float32 [510, 256] .bin files.

Every stage runs with .cpuAndNeuralEngine compute units (the KokoroANE default), so synthesis submits no Metal work and keeps running while the host iOS app is backgrounded. Validation against the PyTorch reference: waveform corr 0.772, log-mel corr 0.994; ~31× real-time on an Apple Silicon Mac under the CPU+ANE-only policy.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for jonschneider/kokoro-coreml

Finetuned
(35)
this model