The useful engineering problem is not “split a song.” A mixed file, or a track already sitting in a library, has to be partitioned into a known set of stems, inspected before anyone keeps the files, written out in a format a later session can open, and handed off so that every file shares the same length. If any one of those steps is vague, the folder you download is just a pile of clips. This is a practical walkthrough of how one public tool page states each of those steps. It is not a quality ranking, not a first-person trial, and not a pricing review. I am not claiming that the page uses any particular source-separation model; the subject is the published contract for inputs, plans, preview, export, and alignment.
What you can feed the page, and what you cannot assume from the filename
The page in question is Stem Splitter. The heading is Stem Splitter. Uploads can be MP3, WAV, FLAC, M4A, AAC, OGG, or WMA, and the page also lets you pick a song from the SongMaker library, including songs made with the AI Song Generator. The upload limit printed on the page is ten minutes; an official example file is named original-take1.wav and is marked 03:16, under the caption “up to 10 min.” You need an account to split songs.
That input list is broader than “a WAV from a session.” A compressed M4A and a lossless FLAC are both accepted. The later export step will not automatically inherit the input codec, so the upload format and the download format are separate decisions. The library path matters for a different reason: some of the files you might split were not recorded in a DAW at all; they were generated in the same studio and already live in the library. The page treats those library songs as valid sources. It does not, on the published text, treat the filename or the origin of the file as a substitute for choosing a plan.
The plan is an enum, not a quality slider
The page asks you to choose a separation plan, and the labels are explicit.
- Vocals & Instrumental · 2 stems — quickly separate vocals and instrumental.
- Standard Split · 4 stems — vocals, drums, bass, other.
- Detailed Split · 6 stems — vocals, drums, bass, guitar, keys, other.
- Full Split · 12 stems — a comprehensive multi-stem export, described as twelve stems for the most detailed breakdown.
- Custom Stems — choose individual stems. In Custom mode you pick exactly which instruments to separate, from kick and hi-hat to violin, sitar and choir. The page prints “Pick From 81 Instruments,” “81 instruments in 9 groups,” and “Add stems later.” The FAQ’s shorter version is that Custom lets you choose individual instruments from dozens of options, such as kick, snare, or backing vocals.
Those five options are not five levels of the same output. Two stems is a vocal-remover contract: the FAQ says you can download the instrumental for karaoke or covers, mute the vocals in the mixer and export a mixdown, or download the vocals as an a cappella. Four stems already takes drums and bass out of the instrumental bucket. Six stems adds guitar and keys. Twelve stems is a fixed high-count breakdown. Custom is a shopping list, with the option to add stems later rather than request every part on the first pass.
The page also prints how the plans are metered and how long they are said to take. A 2-, 4-, or 6-track split costs four credits for any length and usually takes one to six minutes. The 12-stem Full Split and Custom are priced by song length — Custom also by the stems you pick — and take about three to eight minutes. You can start with ten free credits; the FAQ says a four-credit 2/4/6 split means those free credits cover two splits, and that Full Split and Custom are not covered by the free credits. I am repeating those figures only as official page text, not as independent data or a review of price or speed. The engineering takeaway is simpler: 2, 4, and 6 share one metering and time band; 12 and Custom share another. Choosing twelve because “more stems must be better” is not a choice the page asks you to make. It asks you to pick a bundle.
Listen before the files leave the page
The published sequence is three steps: upload a song or pick one from the library, choose how many stems, then mix and export. The third step is easy to rush. The page is unusually direct about it. Results open in a built-in mixer so you can check every track before you download it. You can solo, mute, set volume and pan for each stem, then export single stems or a custom mixdown. The page invites you to try that mixer with a 6-stem split of a track titled Night Drive. Every split is saved in stem history.
That mixer is not a decorative player. It is the inspection step that belongs after the partition and before the folder hits disk. If a 4-stem split has folded a part you care about into Other, you can hear that before you export. If you only needed the instrumental, you can mute the vocal and export a mixdown without downloading a stem you will not use. If you are heading into a DAW, you can still listen on the page first, then export the stems as files rather than as a bounced mix. The history entry means the inspection does not have to be a one-shot decision in the same sitting.
The work runs in the browser, on desktop or mobile, with no plugin or download required for the splitter itself. That is a deployment detail, not a reason to skip the listen. A browser mixer is still a mixer: solo, mute, level, and pan are the checks the page actually names.
Export format is a second decision, made after the listen
Stems can be downloaded as MP3, WAV, or FLAC, one by one, in a batch, or as a single mixdown. Those are three grouping choices and three codecs, and they should not be collapsed. A mixdown is the right object if the mixer has already done the work — for example, a karaoke bed made by muting vocals. A batch of stems is the right object if the next environment is a multitrack session. A single-stem download is the right object if you only need the a cappella or only need the kick from a Custom split.
For a DAW handoff, the FAQ is specific: export WAV or FLAC and drag the stems into Ableton Live, FL Studio, Logic Pro, or any other DAW. Each stem keeps the song’s full length, so the tracks line up. That full-length rule is the alignment contract. A guitar that enters at 0:45 still lives in a file that starts at 0:00 and runs to the end of the song. If the files were trimmed to the first audible frame of each part, you would be left to line them up by hand.
This is ordinary stem craft, not a feature unique to one product. Ableton’s official help article Importing and exporting stems makes the same length requirement from the other direction. When you export stems from Live, the article says to make sure Render Start and Render Length span the entire length of the arrangement, from the beginning to the very end, regardless of where an individual instrument begins or ends. When you import stems, it tells you to turn off Auto-Warp Long Samples so that start markers line up at the beginning of the audio files, even if there is silence at the start. The article also reminds you that, after a drop-in, all the stem audio files should be the same length. I am citing that page as general industry context for why a full-length export matters. I am not saying the SongMaker page uses Live’s exporter, and I am not scoring either product against the other.
If you drag WAV or FLAC stems into Live, FL Studio, or Logic Pro, the useful first check is not “does this sound like the mix.” It is “do these files all start at the same place and run for the same duration.” The official splitter FAQ already says they should. The Ableton article explains why a session falls apart when they do not.
A recording remains a recording after it is partitioned
The page will accept an upload or a library track. That acceptance is not a rights decision. The U.S. Copyright Office, in Circular 56, Copyright Registration for Sound Recordings, states that the copyright in a sound recording covers the recording itself, not the music, lyrics, or other underlying content embodied in it, and that a recorded performance and the musical work being performed are distinct works. The circular also notes that a phonorecord includes digital formats such as .mp3 and .wav files. If the mix belongs to someone else, splitting it into two stems or into twelve does not, on its own, replace those rights. This walkthrough is not legal advice, and it is not a claim about the legal status of the tool. It is only to keep the engineering steps from being mistaken for a clearance step.
The marketing surface in front of all of this is SongMaker, which the site describes as an AI music generator that turns a line of text into a full song, and as a place to create, edit, extend, and split music in one studio. That homepage is a different job from the splitter page. It tells you the studio can begin with a line of text and can later host a split. It does not choose a stem count, an export codec, or a DAW handoff for you. Those decisions belong on the tool page, in the order the page already prints: get a source in, pick a plan, listen in the mixer, export in a format the next session can use, and keep every stem the length of the song so the tracks line up.

Top comments (0)