Stems

Stems ST-02 is a focused utility for audio source separation, built around Facebook’s Demucs v4 model. It does one thing—splitting a mixed track into vocals, drums, bass, and other—and does it with a level of quality that marks a clear generational leap over earlier tools like Spleeter. For DJs, producers, and music learners who need clean, fast stem extraction without the overhead of a full DAW, Stems ST-02 offers a streamlined, no-nonsense workflow. But its simplicity is both a strength and a constraint, and understanding where it fits—and where it doesn’t—is key to getting real value from it.

The headline feature is the separation engine itself. Demucs v4, developed by Facebook Research, uses a hybrid architecture combining convolutional and transformer-based models to achieve noticeably cleaner isolation with fewer artifacts and less bleed than previous versions. In practical terms, that means vocals sound more natural and less phasey, drums retain punch without excessive cymbal wash, and bass lines stay defined rather than muddying into the “other” stem. For a producer pulling a vocal sample for a remix, or a DJ building a live set with isolated drum tracks, this quality difference directly translates to less cleanup work and more usable output. The tool also includes key and BPM detection, which is a smart addition for DJs who need harmonic mixing data and producers who want quick tempo reference. In testing across pop, rock, and electronic genres, the detection was reliable, though it occasionally struggled with tracks that have ambiguous tonality or extreme tempo changes.

The interface is deliberately minimal: upload an audio file, wait for processing, download the four stems as WAV files. There are no batch processing options, no adjustable parameters, no preview before export. This is fine for one-off tasks but becomes tedious for users processing multiple tracks. The naming convention—vocals.wav, drums.wav, bass.wav, other.wav—is clear but rigid; users who need custom file names or folder organization will have to rename manually. More critically, the tool only outputs WAV format, which is lossless but large; MP3 or AAC export would be welcome for users who need smaller files for quick playback or sharing. The lack of finer granularity (e.g., separating guitar, piano, or synth) is a genuine limitation for advanced producers, but it’s consistent with the tool’s positioning as a utility rather than a full production suite.

Who benefits most? DJs will find the key/BPM detection and clean drum/vocal isolation immediately useful for preparing sets, practicing beatmatching, or creating mashups. Producers working with samples will appreciate the Demucs v4 quality for remixes and covers, especially if they need a clean vocal or instrumental stem quickly. Music learners can isolate bass lines or drum patterns to study and replicate, though the lack of transcription or notation means they’ll still rely on their ears. The tool is less suited for sound designers who need multi-stem separation with fine control, or for anyone who needs to process large libraries efficiently.

A practical buyer should weigh the quality of Demucs v4 against the tool’s simplicity. If you need occasional, high-quality stem extraction and don’t want to wrestle with complex software, Stems ST-02 delivers. If you need batch processing, more stem types, or export flexibility, you’ll hit its limits quickly. Pricing details are not provided here, so cost is an open question—but given that free Demucs-based tools exist (e.g., via GitHub), the value proposition will depend on how much you value a polished interface and the added key/BPM detection. For now, Stems ST-02 earns its place as a capable, specialized tool for focused separation work, but it’s not a one-size-fits-all solution.