kokoro constant

DownloadAsset const kokoro

hexgrad's Kokoro 82M voices for Text to Speech, as Google's LiteRT community converted them (int8, with an FP16 vocoder for the GPU delegate; English, Spanish, French, Italian, Portuguese, Hindi, Chinese and Japanese at 24 kHz): 295 MB, Apache 2.0.

This is the file Google's guide uses, with the GPU delegate. Its model card warns that on the CPU delegate, or with no GPU accelerator, its vocoder runs on the CPU and the speech, though intelligible, may be noisier, and names kokoro_int8_fp16.litertlm for CPU-only devices. This package offers Kokoro the GPU delegate only on macOS 15 and later on arm64; elsewhere it runs on the CPU. That file is not pinned: pass a modelPath to a copy of it to load it.

Implementation

static const kokoro = DownloadAsset(
  url:
      'https://huggingface.co/litert-community/Kokoro-82M/resolve/'
      '33ad69fec6c4d9eb10a547aaf9a8c63e7c446a56/'
      'kokoro_int8_gpu_fp16.litertlm',
  sha256: '1ef178b127ec48c737daf17eabb4cea5f6b3ffbc510aced270fe5c267bbeac52',
);