Separate vocals and accompaniment: listen, check, transcribe

What separation gives you

A finished mix combines a lead vocal, backing singers, instruments, room effects and dynamic processing. An AI model estimates which parts belong to vocals and which belong to accompaniment. This is not the same as recovering the original studio multitrack. Reverb may remain with the vocal or leak into the backing track, and a quiet instrument can be mistaken for part of the voice.

For recordings you own or are authorised to use, a backing track can support singing practice or a cover project. A vocal stem can make phrasing, melody and articulation easier to hear. Teachers can compare a part with the full mix, while musicians preparing a transcription can focus on one sound at a time. Check the permission for that intended use: access to a recording alone does not establish the right to distribute a derivative version.

If you want to study the bass line, piano or drums separately, continue with multi-stem separation. The two-stem mode groups the backing instruments together; it does not turn each instrument into its own track. Choose the task that matches your listening goal before processing, then judge the result against that goal rather than expecting a replacement for the original studio session.

Set realistic expectations

Heavy reverb spreads vocal energy across the mix. Choirs and layered backing vocals make the boundary between lead and background less clear. Wide stereo effects and instruments sharing frequencies with the voice can also confuse the estimate. You may hear vocal bleed, watery artifacts in the accompaniment, or softened transients on drums and plucked instruments. Results vary substantially between songs; perfectly clean separation is not guaranteed.

Low-bitrate files, repeated lossy encoding and aggressive compression can remove information before separation starts. Use a higher-quality authorised source when available, and avoid unnecessary re-encoding. Exporting a damaged source to a lossless format does not restore the missing detail. Listen to the sections you plan to practise or use, including busy passages, rather than deciding from a quiet introduction alone.

From import to export

Prepare your own recording, a work you created, public-domain material or audio you have explicit permission to process. Open StemTab and import the audio, then choose the “Vocals & accompaniment” mode. The Web studio uses cloud GPU processing; the iPhone and iPad App processes projects on device by default. Processing time depends on the file, task and device or service conditions, so this guide does not promise a fixed turnaround.

After processing, listen to the vocal and accompaniment stems separately and compare them with the original mix. Check entrances and exits, stacked backing vocals, strong drum hits and the ending. These transitions often reveal missing sounds or unwanted remnants. Keep playback levels moderate and comparable: a louder stem can seem clearer even when its separation is not better. Select the parts you want to hear and check the mix before exporting.

Continue with multi-stem separation if you need vocals, drums, bass, guitar, piano and other parts individually. To work on notation, transcribe the relevant part into an editable score draft. Listen while correcting pitch, duration, rests and rhythm. Dense chords and short notes deserve particular attention. Automatic transcription is a starting point for revision, not a finished score ready to publish without checking. StemTab does not generate guitar tablature (TAB).

Export audio for listening or further production, PDF for reading and printing, MIDI or MusicXML for continued editing, and a .stemtab project package to retain a project. Check whether you want individual stems or a mix of selected parts before exporting. The appropriate format depends on the next job: a printable page and an editable sequence serve different purposes, even when they come from the same transcription.

A practical listening and revision check

Listen during vocal rests for traces of lyrics in the backing track. When the singer enters, check whether an instrument disappears or loses body. In dense passages, listen for unnatural movement in cymbals and sustained instruments. Check the start, ending and section changes as well as the chorus. Headphones and familiar speakers can reveal different problems, so compare them when the quality matters for your intended use.

If a result has obvious artifacts, first check whether distortion is already present in the source and whether a better-quality authorised version exists. A different export container will not resolve heavy reverb or overlapping parts. Treat imperfect stems as listening and editing references when appropriate. Quality review and permission to use the recording are separate checks; passing one does not establish the other.

For notation, confirm the beat structure and section boundaries before reviewing dense chords, fast runs and note endings. MIDI represents note events; MusicXML carries notation structure for further editing; PDF provides a reading and printing document. None replaces musical judgement. After editing, listen again and check that the written rhythm is readable and matches what you intend to practise.

Rights and billing

Only process audio you own or are authorised to use, within the scope of that permission. Being able to import a file does not grant rights to arrange, perform or publish it. StemTab does not supply songs or sheet music for download.

Web billing is based on processing minutes. Consult the pricing page for current rules before starting a task. App purchases and Web usage follow their respective product terms; this guide does not quote package prices or allowances.