Convert .pth voice models to .onnx format right in your browser. No upload, no Python, 100% private.
Browser-based · No download requiredDrop your .pth model file into the converter
The converter analyzes the model structure
Your model is converted to ONNX format locally
Download the .onnx file and import into Echo
Your model file never leaves your computer. All conversion runs locally in your browser using WebAssembly — no server, no cloud, no data collection.
No Python, no PyTorch, no command line. Drop your .pth file and get a ready-to-use .onnx model in seconds. Works on any modern browser.
Supports all RVC v2 voice models (768-dim embedding) from weights.gg, HuggingFace, AI Hub, and Applio. Output works directly with EchoRVC.
Convert character voice models from weights.gg to use with EchoRVC during gaming sessions on Discord, Valorant, or CS2.
YouTubers and streamers can convert celebrity or character voice models for comedy skits, impressions, and live entertainment.
Convert vocal RVC models to create AI covers, backing vocals, or vocal style transfers in your DAW workflow.
ML researchers and hobbyists can quickly convert models between formats for testing, benchmarking, and cross-platform deployment.
RVC (Retrieval-based Voice Conversion) is the most popular open-source voice cloning technology. Most community-trained models are shared as .pth files — PyTorch checkpoints that require a full Python environment with CUDA and PyTorch installed to run. This is a significant barrier for non-technical users who just want to change their voice. ONNX (Open Neural Network Exchange) is a portable model format that runs on any platform without Python. Converting your .pth model to .onnx makes it compatible with EchoRVC, our native desktop voice changer that achieves real-time inference without any Python dependencies. The result is faster loading, lower latency, and a seamless plug-and-play experience.
Until now, converting RVC models required running a Python script with PyTorch installed — a process that trips up even experienced users. Other online converters require you to upload your model file to a server, raising privacy concerns and introducing upload wait times for files that can exceed 100 MB. Our converter is different. It uses the rvc-onnx-web library to perform the entire conversion client-side using WebAssembly. Your model file stays on your machine from start to finish. There is no upload, no server processing, and no data retention. This makes it the most private and accessible way to convert RVC voice models anywhere on the internet.
Yes — conversion happens 100% in your browser. Your .pth file is never uploaded to any server. You can verify this by checking the Network tab in your browser's developer tools — zero requests are made during conversion.
PTH files are PyTorch checkpoints that require a full Python environment to run. ONNX is a portable, optimized format that runs on any platform without Python. EchoRVC uses ONNX models for maximum speed and compatibility across Windows, macOS, and web.
This converter supports RVC v2 models with 768-dimensional embeddings, which is the modern standard used by Applio, Mangio-RVC, and most community trainers since 2023. Older RVC v1 models (256-dim) are not supported — use Applio's built-in ONNX export for those.
Change your voice online for free. Record your mic or upload audio, pick a voice preset, and download instantly. Chipmunk, robot, demon, Darth Vader, and more.
Record your voice online for free. No download, no signup. Browser-based voice recorder with instant WAV/MP3 export. Works on any device.
Convert audio files between WAV, MP3, OGG, and FLAC for free. Browser-based, no upload required. 100% private.
Remove background noise from audio files for free. AI-powered noise reduction processes everything in your browser. Clean up recordings, podcasts, and voice memos instantly.
Drop your .pth model here
or click to browse · RVC v2 models up to 200 MB
Supports .pth files from weights.gg, HuggingFace, AI Hub, and Applio
100% client-side · Zero uploads · All processing happens in your browser