Start with a clean source
Use the highest-quality recording you have permission to process. Lossy compression, distortion and heavy reverb can make instrument boundaries harder for an AI model to identify.
Choose the stem groups you need
Audixo downloads a Demucs AI model on first use, then estimates Lyrics/Vocals, Music, Drums, Bass and Other Instruments on your device. These are model predictions rather than the original session tracks. Keep input under six minutes and 120 MB. The main singing voice and backing music are separated first, while individual singers and every original instrument may still contain bleed or artifacts.
Audixo noteMake one change, listen, then decide. Clear A/B decisions beat stacking effects blindly.
Listen to each stem in context
Solo a track to find bleed, metallic artifacts or missing transients, then listen again with the other tracks enabled. Small artifacts are sometimes less audible in the combined mix.
Keep the timing aligned
The separated tracks start together. When removing a middle section, cut all synchronized tracks at the same times to keep the rhythm and vocal phrasing aligned.
Save the useful outputs
Keep the returned stems in the Audixo editor, mute or replace any layer, and add a new beat, keys, guitar, synth, bassline or transition from the separate Music & Beats library. Download important WAV/MP3 results before clearing browser data. Rights in the original recording still apply.
Now make it yours.
Bring a real recording into Audixo and use the workflow while it is still fresh.
Open Studio ↗