CPU-only multilingual text-to-speech process extension for Modly using Kokoro ONNX.
This extension integrates thewh1teagle/kokoro-onnx with Modly to convert text into mono PCM16 WAV audio at 24 kHz. The Modly wrapper is authored by DrHepa and runs inference through ONNX Runtime's CPU provider.
1. In Modly, open Models/Extensions → Install from GitHub. 2. Enter https://github.com/DrHepa/modly-kokoro-onnx-extension. 3. Wait for setup to create the extension venv, install pinned dependencies, download both official Kokoro v1.0 assets, and verify their integrity.
The extension pins kokoro-onnx==0.5.0, onnxruntime==1.28.0, and numpy==2.4.6, and provisions Kokoro v1.0 from the upstream model-files-v1.0 release bundle.
Process extensions do not have Modly's model-weight download button. Install and Repair therefore own asset provisioning. A valid existing asset is never downloaded again, and generation never uses the network.
The standard installed layout derives models beside extensions. Current process/setup payloads do not carry a custom models directory, so custom Modly layouts must provide the same absolute MODELSDIR environment variable to both Install/Repair and generation.
Read the full README on GitHub →