supertonic and TensorVox
These tools are competitors, as both aim to provide on-device neural speech synthesis, with Supertonic offering a more established and widely adopted solution utilizing ONNX for cross-platform deployment, while TensorVox is a less mature, C++ native desktop application.
About supertonic
supertone-inc/supertonic
Lightning-Fast, On-Device, Multilingual TTS — running natively via ONNX.
Supports voice customization through Voice Builder for creating edge-deployable speaker profiles, and provides multi-runtime SDKs (Python, Node.js, C++, Swift, iOS, Flutter, etc.) enabling deployment from Raspberry Pi to browsers via WebGPU/WASM. Achieves extreme performance with just 66M parameters and configurable inference steps, reaching 167× real-time factor on consumer hardware while maintaining complete on-device privacy without cloud dependencies.
About TensorVox
ZDisket/TensorVox
Desktop application for neural speech synthesis written in C++
Supports multiple ONNX and legacy TensorFlow/LibTorch models (VITS, Tacotron2, FastSpeech2, Supertonic) with GPU acceleration via DirectML on Windows, enabling efficient on-device inference. Built with Qt and ONNX Runtime as the primary backend, it handles phoneme-based synthesis across English, German, and Spanish out-of-the-box, with extensible language support via IPA/ARPA phoneme sets. Includes integrated audio post-processing with RNNoise denoising and r8brain resampling, plus experimental Logitech RGB peripheral integration.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work