Can RVC run on a Mac?
Yes. RVC was built around NVIDIA graphics cards and CUDA, which Macs do not have, but Apple Silicon Macs (M1 and newer) can run it through Apple's own GPU framework instead. Conversion works well. Training works too, it just tends to take longer than on a high-end NVIDIA card, so plan your runs accordingly.
Intel Macs can technically run the Python versions, but they are slow enough that most people give up. If you have an Intel Mac, expect long waits or use a cloud option.
Option 1: the original RVC WebUI
The RVC Project's WebUI is the original tool. You install Python, ffmpeg, and the project's dependencies, download the pretrained models, then launch a local web page in your browser. It gives you every setting, including training, model merging, and checkpoint extraction.
The cost is setup time. Version mismatches between Python packages are the most common way installs fail, and most troubleshooting threads assume Windows. If you enjoy the terminal and want full control, it is the reference implementation. If you just want to convert a vocal, it is a lot of work for the first result.
Option 2: Applio
Applio is a free, open-source app built on RVC and released under the MIT license. Its desktop app has builds for both Apple Silicon and Intel Macs running macOS 12 or later, which removes most of the Python setup. It covers conversion and training and includes extras such as text to speech and a model search.
One thing to know: Applio's download page says the Mac app is not Apple-notarized yet, so macOS may report it as damaged on first launch. Their page gives a Terminal command to clear that flag. We compare the two in more detail in Applio vs RVC.
Option 3: Xport Studio
Xport Studio is a desktop app for Mac and Windows that runs RVC voice conversion, training, and text to speech locally, alongside stem splitting, key and BPM detection, noise removal, and mastering. The Mac build is signed and notarized by Apple, so it opens like any other app, and it needs no Python.
To use a voice you already have, drag the .pth file into the Library and convert a vocal with the vocal converter. Voice conversion is free for 7 days, then part of Xport Pro, $79 one time. It needs macOS 15 or later on Apple Silicon; older macOS versions are not supported.
Training a voice on a Mac
Training is the heavy part. On Apple Silicon it works, but runs take longer than on a fast NVIDIA card, and the fans will spin. Keep the Mac plugged in, close other heavy apps, and start with a modest dataset: 10 to 20 minutes of clean, dry vocals is a good first project.
Our guide on how to train an RVC model covers how much audio you need and how many epochs to run. In Xport Studio, Train Clone handles the whole run on your Mac, and a typical session finishes in 30 to 90 minutes; see voice cloning for details.
Common Mac problems and fixes
"App is damaged and can't be opened": macOS shows this for apps that are not notarized by Apple. It usually means the app has not been through Apple's check, not that the download is broken. Only clear the flag for apps from a source you trust, using the instructions the developer publishes.
Running out of disk space: RVC tools download pretrained base models on first run, which can take several gigabytes, and every trained voice adds roughly 60 MB plus its index. Leave plenty of free space before you start a training run.
Slow or stalled training: close browsers and other GPU-heavy apps, keep the Mac on power, and try a smaller batch size if the tool exposes one. If a run seems frozen, check the log before restarting; on most tools the progress just updates in bursts.
Tips for better results on any Mac setup
Feed RVC a dry vocal. Reverb, backing vocals, and instruments confuse the model and come out as warble. If your vocal is part of a full song, separate it first with the stem splitter.
Set the pitch to match the voice: leave it at 0 for a similar range, raise it 12 semitones for a male vocal into a female model, or lower it 12 for the reverse. And keep each model's .pth and .index files together in one folder so you never have to guess which index belongs to which voice.