OpenVoiceChanger
github.com
- Free plan
- Windows
- Linux
- Web
- API
- Self-hosted
What OpenVoiceChanger is
OpenVoiceChanger is a free, open-source, self-hosted web application for changing voices in real time or from recorded files. It streams microphone audio over WebSockets and supports ONNX, RVC, and model-free DSP modes. The studio provides pitch and formant shifting, 16 F0 methods, 12 DSP effects, and 16 built-in voice presets, with custom presets saving complete Voice Lab settings. Its offline converter accepts WAV, MP3, FLAC, OGG, and M4A files and renders WAV output, up to a configured maximum of 600 seconds. ONNX models and RVC PyTorch .pth or .pt models are supported; RVC requires a HuBERT content encoder. HTTP endpoints cover health, models, presets, and conversion, while a WebSocket endpoint handles live audio. The browser application supports Linux, Windows, and self-hosted deployment. Requirements include Python 3.10 or later, Node.js 20.19 or later, and npm. The MIT-licensed project stores recorded takes in browser IndexedDB, where the browser may clear or evict them.
Who it suits
It suits users who want local, browser-based voice conversion and audio effects. Developers can also use its HTTP and WebSocket endpoints.
Good to know, and what to watch
- Free and open source under MIT
- Supports real-time and recorded-audio conversion
- Includes ONNX and RVC model support
- Offers 12 DSP effects and 16 presets
- Runs self-hosted on Linux or Windows
- Recorded takes may be cleared by the browser
- RVC models require a HuBERT encoder
- Offline conversion is capped at 600 seconds
- Requires Python, Node.js, and npm
The verdict
OpenVoiceChanger offers local real-time conversion, file processing, model support, and configurable effects without a paid plan. Its self-hosted requirements and browser storage behavior are important considerations for setup and recordings.
What the maker publishes
- API
- The backend exposes HTTP endpoints for health, models, presets, and conversion, plus a WebSocket endpoint for live audio. source
- API
- The backend exposes HTTP endpoints for health, models, presets, conversion, and updates, plus a WebSocket endpoint for real-time audio. source
- Checkpoint security
- Uploaded RVC checkpoints use PyTorch safe weights-only loading by default; unsafe loading requires an explicit deployment setting. source
- Custom presets
- Users can save custom presets containing the complete Voice Lab settings. source
- Deployment
- The app runs a local backend and browser UI, with the backend bound to 127.0.0.1 by default. source
- Effects
- Its server-side DSP rack includes 12 effects, including noise gate, distortion, chorus, echo, reverb, EQ, and compression. source
- File conversion
- The offline converter accepts WAV, MP3, FLAC, OGG, and M4A files and renders WAV output; the configured maximum is 600 seconds. source
- File conversion
- Users can upload WAV, MP3, FLAC, OGG, or M4A files for conversion to downloadable WAV, with a configurable maximum length of 600 seconds. source
- Integrations
- The README documents ONNX Runtime, PyTorch, RVC, and GitHub Releases for updates; it does not list third-party conferencing integrations. source
- License
- The repository identifies its license as MIT. source
- License
- The repository is distributed under the MIT License. source
- Live conversion
- It streams microphone audio over WebSockets and also supports model-free DSP voice effects. source
- Local storage
- Recorded takes are stored in browser IndexedDB on the same origin and may be cleared or evicted by the browser. source
- Model formats
- The app supports ONNX models and RVC PyTorch .pth or .pt models, with RVC requiring a HuBERT content encoder. source
- Models
- It accepts ONNX and RVC .pth or .pt models, and RVC models require a HuBERT content encoder file. source
- Network security
- Cross-origin HTTP and WebSocket requests are checked against an origin allowlist, and the README recommends HTTPS for LAN microphone access. source
- Platforms
- The quick-start instructions target Windows PowerShell; the README says other platforms use manual setup and documents POSIX-shell or WSL Makefile commands. source
- Platforms
- The README documents Windows setup and says other platforms use manual setup; its Makefile supports POSIX shells or WSL. source
- Product
- OpenVoiceChanger is a self-hosted web application for real-time AI voice conversion using ONNX or RVC models. source
- Purpose
- OpenVoiceChanger is a real-time AI voice changer web application that routes a microphone through ONNX or RVC models. source
- Realtime conversion
- It streams audio over WebSockets and supports ONNX, RVC, and model-free DSP modes. source
- Requirements
- The listed requirements include Python 3.10 or later, Node.js 20.19 or later, and npm. source
- Requirements
- The documented requirements include Python 3.10 or later, Node.js 20.19 or later, and npm. source
- Security
- The backend binds to localhost by default, checks browser origins for HTTP and WebSocket requests, and loads RVC checkpoints in safe weights-only mode by default. source
Checked against the maker's pages on 9 Oct 2026.