Ever wanted to hear your demo sung in a different voice, or pull the vocals out of a song for a remix? Replay, a free desktop app from the makers of Weights, runs AI voice models (RVC) and stem splitting locally on your PC.
Replay applies voice models to vocal recordings
Add an audio track or record vocals through a microphone, then choose a voice model from the Weights community library. Replay uses the selected model to change the vocal sound, letting creators try different voices on a song or their own recording.
The library holds thousands of community-made models, and you can load your own trained model files. Using a real artist's voice without permission can raise copyright and personality-rights issues, so keep such results for private experiments.
Separate vocals and instruments from a track
Replay can split an audio track into vocal and instrumental parts. This gives remixers and producers separate material to work with, such as extracting a vocal for a new arrangement or isolating instruments from the original track.
- Import a song to separate its vocal and instrumental content.
- Use the isolated parts as material for a remix or alternate version.
Work with audio without an internet connection
Processing runs on your own hardware, so no audio is uploaded; a recent NVIDIA graphics card speeds it up a lot, while CPU-only conversion is slow. The internet is only needed to download models. Its core workflows are loading or recording vocals, applying a voice model, and separating track elements.
Pros
- Applies community voice models to supplied or recorded vocals
- Separates vocals and instruments from audio tracks
- Supports offline audio work
Cons
- Slow without a capable NVIDIA GPU
- Using real artists voices can raise legal issues
- Large model downloads
System requirements: Windows 10, Windows 11
Can Replay change the voice in an existing song?
Yes. It first separates the vocal, converts it with the chosen model, then mixes it back with the instrumental.
Is Replay free?
Yes. The desktop app is free and runs models locally.
Do I need a powerful PC?
A recent NVIDIA GPU is strongly recommended; on CPU alone, conversions take much longer.
Can I publish songs made with famous voices?
Be careful: using a real singer's voice without consent can break copyright and personality-rights laws and platform rules.