An ONNX-optimized version of the Kokoro 82 million parameter text-to-speech model, enabling efficient cross-platform inference.