A pianist sends you a rehearsal recording made on a phone propped against a music stand. The choir is loud, the piano is louder, and you want one thing out of it: a version you can sing along to at home without the piano fighting your pitch. Before you reach for any tool, it helps to understand what a two stem separator actually promises — and what it cannot deliver from a source like that.
What "Two Stems" Really Means
A two stem separator divides a recording into two outputs: vocals and accompaniment. That is the whole contract. It is not the same as pulling out drums, bass, guitar, and keys as separate files — that is a multi-instrument job, which is a different task entirely.
ShengXi Vocal Remover works within that two-part model. You import a video from Photos or a local audio file, the app separates vocals and accompaniment, and you export the result. Extraction, separation, and export all happen on the device, and your source file does not need to be uploaded to a server.
So when someone asks for "just the piano part," the honest answer is that a two stem tool can hand you the accompaniment — piano plus whatever else was playing. If a second instrument was in the room, it stays in the accompaniment.
Three Questions to Ask Before You Separate
1. Is the voice actually in the center of the mix?
Separation tends to work best when the vocal sits clearly in the middle of a stereo recording and the backing sits wider. A phone recording of a live room often breaks that assumption. If the singer stood beside the piano rather than in front of a microphone, the tool has less to work with.
2. How much noise is baked in?
Room hum, page turns, chair scrapes, traffic, and audience chatter all get treated as part of the material. The separator does not know a cough from a chord. Expect the rough edges of your source to survive the process, sometimes more noticeably than they did in the original.
3. What will you do with the output?
If you want a rough reference to hum along with in the car, a mid-quality split is fine. If you plan to sing over the accompaniment in a lesson, small artifacts become much more distracting. Match your expectations to the purpose.
Choosing Between Fast and Fine Processing
Two modes exist for a reason, and picking the wrong one is a common source of frustration.
- Fast mode — useful when you are testing whether a source is even workable, or when you need a quick answer before deciding whether to invest more time.
- Fine mode — better suited to the take you actually intend to keep, where you are willing to wait for a cleaner result.
A sensible habit: run the fast pass first on anything unfamiliar. If the accompaniment still sounds smeared or the vocal still carries obvious backing bleed, a longer pass probably will not rescue it. The problem is the source, not the setting.
The Preview Step Most People Skip
Once separation finishes, you get a mixed preview plus separate volume controls for the vocal and the accompaniment. This is where you decide what the file is for.
- Pull the vocal all the way down and listen only to the accompaniment. Does it hold together as a backing track, or does it collapse into a hollow, watery version of the original?
- Pull the accompaniment down and listen to the vocal alone. Is it clean enough to use as a guide, or does it carry the piano with it?
- Set a balance you can actually rehearse with, then export that version rather than the extremes.
Many disappointing results come from exporting the maximum-separation version instead of the balanced one. A little bleed often sounds more musical than a perfectly isolated but thin and lifeless stem.
When the Result Is Not Good Enough
The most important thing to accept is that results vary with source quality, mixing, noise, and compression. No tool can guarantee identical or perfect separation, and anyone promising that is overselling.
Before you blame the app, check the source:
- Was it recorded in a quiet room, or at a party?
- Is it a clean stereo file, or a heavily compressed clip pulled from a video?
- Is the vocal mixed forward, or buried under a loud band?
- Was the original mono? Mono sources give a separator far less to work with.
If two or three of those answers are unfavorable, your realistic option is a better recording — a quieter space, the singer closer to the mic, the backing further back. That is not a workaround. It is the actual fix.
Exports, History, and Housekeeping
Finished results can be shared and kept in recent history, so you do not need to redo a separation you already like. It is still worth cleaning out old files periodically — rehearsal material piles up fast, and most of it was only ever a rough reference.
One more practical note: you need permission to process and publish whatever you feed in. The app does not grant rights to music or video, and separating a track does not make it yours to distribute.
Used within those limits, a two stem separator solves a real problem: it turns a single recording into something you can practice against. Just be honest about the source before you expect miracles from the output.