This software processes audio files through a series of machine learning models that identify and separate different frequency ranges and timbral characteristics within a song. Users can upload files directly in MP3, WAV, FLAC, AIFF, MP4, MOV, or M4A formats, or paste a YouTube video link to extract audio automatically. The system then routes the file through one of several specialized separation models depending on what the user wants to isolate.
The separation architecture offers multiple model configurations. A 2-track model splits audio into vocal and instrumental components, or can target specific elements like reverb removal or lead versus background vocals. there is also a 3-track model that divides files into speech, music, and effects categories. The 4-track model isolates bass, piano, drums, and guitar simultaneously. The most complex 6-track model extracts all four instruments plus separate vocal and instrumental stems in a single pass.
Beyond the standard vocal and instrumental split, this software includes specialized models for isolating individual instruments. You can extract only drums, bass, piano, guitar, organ, strings, or wind instruments from a mix. There's a multi-singer model that attempts to separate different vocalists, and a male versus female vocal separator. The reverb removal model strips spatial effects from recordings, which is technically difficult since reverb occupies the same frequency space as the dry signal.
The separation happens in one click after model selection. The system processes the file and provides a preview before download. The underlying models get updated regularly, which means separation quality can improve over time as the algorithms are retrained on larger datasets.
Files can't exceed 20 minutes in duration or 30MB in size. These constraints likely exist because longer files require more computational resources and processing time. The 30MB limit means high-bitrate files might need compression before upload, which could affect the quality of the source material going into the separation process.
This software works across web browsers, iOS, Android, and macOS. This cross-platform approach means the processing happens server-side rather than on the device, so performance won't vary based on local hardware. The YouTube integration is the only external service connection, allowing direct extraction from video URLs without manual downloading and re-uploading.
Pricing starts at $5.99 monthly or $71.99 annually for the Basic plan, which includes 30 stem splitter credits per month. The Pro plan costs $8.33 monthly when billed annually at $99.99, providing 100 stem splitter credits. The Enterprise tier runs $24.99 monthly or $251.88 yearly with 200 credits. There's a free plan offering 5 stem splitter credits, but it doesn't allow song downloads and puts users in a wait queue rather than priority processing. Free users also get credits for other features like AI music generation and cover creation, though those fall outside the vocal removal functionality.
The credit system means each separation consumes one credit regardless of which model configuration you choose. This differs from services that charge based on file length or number of stems extracted. The free tier's download restriction is a significant limitation since you can preview separations but can't save them for actual use.
The lossless instrumental track production suggests the output maintains the original file's bit depth and sample rate where the format allows. Distortion-free processing is the goal, though separation quality depends heavily on the source material's mix complexity and the specific model used.
