Turn any song into a practice track: vocals out, click on, key shown

StemX now turns a song into something you can practise with. Upload a track and you get the instrumental without the singer, a click locked to the song's own beats, and its tempo and key on screen. It all runs in the browser, so there is nothing to install and no graphics card needed. This post covers what is new, how to use it, and how accurate each part is.

What's new

• A click track from the song itself. A full split now detects every beat and downbeat, and the player can play a click on them alongside the stem you are listening to. Because the click follows the song's own beats rather than a fixed metronome, it stays with recordings that drift.
• Lead and backing vocals. After a full split you can separate the vocals again into the lead voice and the backing vocals. The button shows how many minutes it will use before you press it.
• Tempo and key on every full split. Results show the BPM, with the half- and double-time readings, and the key. The same analysis is free on its own in the BPM finder and key finder.
• Lossless WAV. Any paid plan or minute pack can download every stem as WAV as well as 320 kbps MP3.
• A $4.99 Starter plan with 300 minutes a month, for players who split a few songs a week.

How to make a practice track

1. Open the vocal remover and drop in a song. It accepts MP3, WAV, FLAC, M4A, OGG, AAC and AIFF files, up to 10 minutes and 120 MB.
2. Listen to the free 60-second preview. It needs no account, so you can check the separation on your song before spending anything.
3. Sign in and split the whole song. A free account includes 30 minutes a month.
4. Mute the vocals and turn on the click. Sing over it, or play the melody on your own instrument. The tempo and key above the player tell you what to tune to and where to set your metronome later.
5. Download the stems to keep practising offline, or drop them into your recording software.

How accurate it is

We publish our measurements, including the weak ones:

• Separation. When this post was written, StemX's vocals-and-instrumental model scored a median 11.43 dB SI-SDR on the vocals and 14.15 dB on the instrumental, on 12 tracks from the MUSDB18 test set. We have not yet published the same measurement for the six-stem models StemX runs now. Dense mixes and heavy reverb can still leave traces of the singer. Details of the earlier measurement are in our benchmark.
• Tempo. Within 4% of the real tempo on 81.9% of 998 test tracks. Most misses are half or double the real tempo, which is why the alternatives are shown. See how to find the BPM of a song.
• Key. The key finder answers on about 44% of tracks and is right about 80% of the time when it does. On the rest it says it isn't sure instead of guessing, because a confident wrong key is worse than none when you are about to play along. More in our key detection post.

What it doesn't do

• Six stems. You get vocals, drums, bass, guitar, piano and other, and the vocals can be split into lead and backing. Instruments without a stem of their own, such as strings or synths, land in other.
• Files only. You upload an audio file; StemX does not fetch songs from streaming or video sites.
• No click on previews. Tempo, key and the click come with a full split, not the 60-second preview.

Credited models, deleted uploads

StemX credits every model it runs in the site footer: BS-RoFormer-SW (bs-roformer-infer, MIT; weights enerjazzer/BS-ROFO-SW-Fixed); HT-Demucs 6s (Meta AI Research, demucs MIT) for separation; YourMT3+ (mimbres/YourMT3) for pitched stems and ADTOF (CC BY-NC-SA) for drums; Beat This! for tempo; and S-KEY for key. Uploads are deleted automatically: anonymous previews after an hour, full splits after 3 to 30 days depending on your plan.

Pricing

Only upload music you own or have the rights to use. Practising along to a song at home is one thing; publishing what you make from it is another.

Credit

Separation uses BS-RoFormer-SW (bs-roformer-infer, MIT; weights enerjazzer/BS-ROFO-SW-Fixed); HT-Demucs 6s (Meta AI Research, demucs MIT); the lead and backing vocal split uses UVR models by Anjok07 & aufr33 (MIT, with credit to UVR and its developers). Tempo uses Beat This! by Foscarin, Schlüter & Widmer, CPJKU, and key uses S-KEY by Kong et al., Deezer.

Make your first practice track: the preview is free and needs no signup.