--- license: other license_name: mixed-open-licences tags: [text-to-speech, automatic-speech-recognition, kokoro, whisper, splainer] --- # Splainer voice The voice and speech recognition that [Splainer](https://splainer.shadow-lancer.com) runs on your own PC: free, private, and offline once downloaded. In Splainer: **Settings → Voice → Download**. Splainer checks every file's SHA-256 before using it. | File | Size | What it is | |---|---|---| | `kokoro-v1.0.onnx` | 326 MB | Kokoro voice model | | `splainer-voice-gpu-win-x64.zip` | 1,292 MB | CUDA libraries for NVIDIA cards | | `splainer-voice-win-x64.zip` | 111 MB | the voice program for Windows x64 | | `voices-v1.0.bin` | 28 MB | Kokoro voices | | `whisper-large-v3-turbo/model.bin` | 1,618 MB | Whisper large-v3-turbo | | `whisper-small/model.bin` | 484 MB | Whisper small | - **Every PC** (0.95 GB): the voice program, the Kokoro voice and Whisper *small* (runs on the processor). - **NVIDIA cards with 6 GB or more** (+2.91 GB): CUDA libraries and Whisper *large-v3-turbo* (about 0.2 s per question). ## What is inside, and its licences | Part | Licence | From | |---|---|---| | Kokoro-82M voice (`kokoro-v1.0.onnx`, `voices-v1.0.bin`) | Apache-2.0 | [hexgrad/Kokoro-82M](https://huggingface.co/hexgrad/Kokoro-82M), ONNX export by [kokoro-onnx](https://github.com/thewh1teagle/kokoro-onnx) | | Whisper large-v3-turbo (CTranslate2) | MIT | [mobiuslabsgmbh/faster-whisper-large-v3-turbo](https://huggingface.co/mobiuslabsgmbh/faster-whisper-large-v3-turbo), model by OpenAI | | Whisper small (CTranslate2) | MIT | [Systran/faster-whisper-small](https://huggingface.co/Systran/faster-whisper-small), model by OpenAI | | Voice program (`splainer-voice-win-x64.zip`) | its source is in `source/` | built with PyInstaller from `source/kokoro_server.py`; bundles kokoro-onnx (MIT), ONNX Runtime (MIT), faster-whisper (MIT), CTranslate2 (MIT), PyAV/FFmpeg (LGPL) and espeak-ng (GPL-3.0) | | CUDA libraries (`splainer-voice-gpu-win-x64.zip`) | NVIDIA licences, included in the archive | cuBLAS, cuDNN and NVRTC runtime libraries from NVIDIA's `nvidia-*-cu12` packages | `source/` holds the server's code, the exact package versions and the build script, so the program can be rebuilt. It only listens on 127.0.0.1 and only answers Splainer (a per-install token).