Player not loading? Watch on YouTube
The speaker compares AI stem splitters using 13 songs with different genres and arrangement complexity, judging separation, artifacts, tone and ease of use. These rankings reflect the speaker's listening tests. Dense mixes and stacked vocals proved harder to separate than sparse acoustic arrangements.
Ultimate Vocal Remover (UVR) is the speaker's preferred option for power users. Described as free and open source, it processes audio on your own computer, making it relevant to local AI workflows. The tradeoff is an extra processing step and time spent trying different algorithms. For vocals, the speaker favors a vocal-specific algorithm; for instrumentals, the choice is MDX23 XC Inst Vocal HQ.
Moises earns the recommendation for background vocals and quick upload-based separation. Bass results are less convincing across the tested tools: the speaker reports missing upper frequencies and softened or omitted transients. Drum separation generally fares better, especially with consistent, processed drum-machine sounds, though no algorithm wins for every source.
For people who already own a DAW with stem separation, the speaker recommends starting there. Ableton is the preferred DAW option for sound quality, with a reported speed limitation on Mac. LALAL disappoints relative to alternatives, while RipX and RX 11 receive stronger criticism for separation quality.
