Preparing stems for a live mashup.

Preparing stems for a live mashup

The premise of this workflow is already set out in this topic area: separate and save before the live set, then leave the booth to play prepared WAV files. The guide to live mashups with acapellas, stems and decks makes the point clearly: preparing whatever you can in advance frees up CPU resources and avoids discovering artefacts in front of an audience. This article covers the complete workflow around that idea, from choosing the song to having a backup plan ready.

---

1. Why separate before the live set

Separating stems on the spot uses the same computer needed for playback, makes you wait for the model’s first load and turns every set into a different test: the same mix does not sound the same on two consecutive nights if separation happens live. Working beforehand gives you the same result every time, lets you listen calmly for artefacts on your own system and leaves the CPU free for effects, stems already loaded and whatever else the night calls for.

When live separation is available, it is at most an emergency option, rather than the plan. With pre-rendered stems, you know exactly what you will hear.

---

2. Choosing and trimming the source (steps 1 and 2)

Step 1: choose the intended mashup. Start with a combination that works: the same key or compatible keys, tempos you can match and phrase structures that fit together. This advance selection is covered in the guide to loops, harmony and mashup preparation and the guide to produced mashups: tempo, key and warping.

Step 2: trim the sections first. Before separating, trim the source to what you will actually use: the intro, chorus or window where the acapella enters. As well as simplifying your review, this addresses a hard limit of the separator: the maximum input is 480 seconds, or eight minutes. Split long songs beforehand, rather than afterwards.

---

3. Separation and cleanup (steps 3, 4 and 5)

Step 3: separate the stems. The browser separator works with four stems—vocals, drums, bass and instrumental—using a deterministic Pure DSP engine with 0 MB download. It processes audio directly in your browser with instant, fully private harmonic/percussive spectral separation. Each stem is exported as a 16-bit WAV at the original sample rate.

Step 4: clean up if necessary. If the source contains background noise or reverberation, cleanup comes before separation: apply denoise and dereverb to the original track, and only then separate it. That principle, along with the cautions for each tool, is covered in the guide to AI stem separation, denoise and dereverb.

Step 5: export WAV. Stick with WAV: it is the format you will then work with in the booth, and it does not depend on codecs at the critical moment.

---

4. Naming and tagging (step 6)

Step 6: a naming convention. Name every file using a clear, repeatable structure: track, stem and version, for example tituloinstrumental-v1. With four stems per song and several songs in play, a strict convention keeps you from loading one track’s vocals while thinking they belong to another.

---

5. Beatgrids and cues for the stems (step 7)

Step 7: grids and points for each stem. Import each stem into your software and treat it like any other track: review the beatgrid first, then mark cues and loops where the mashup enters and exits. Stems from the same song share their duration and sample rate, but each one plays and launches separately, so each stem needs its own points. The grid reference is in the beatgrid guide.

---

6. Gain (step 8)

Step 8: consistent levels. Set the gain so that stems of different types arrive at levels that are coherent relative to one another: separated vocals, bass and drums do not automatically retain the volume relationship of the original master. Follow gain-staging and metering principles, as explained in the guide to dBFS, LUFS and gain staging, and leave the channels low enough to raise them in the booth.

---

7. Importing for the booth and soundcheck (steps 9 and 10)

Step 9: a dedicated crate or playlist. Bring all the mashup’s stems together in their own working group, with the same visible naming convention. If the mashup is scattered across your general library, you will be searching at the critical moment instead of performing.

Step 10: soundcheck the exact mashup. Run the complete combination once, with the stems loaded, on the system you will perform with or the closest available equivalent. This is the step that separates “I have the files” from “I have the mashup”.

---

8. A backup plan (step 11)

Step 11: a rendered backup. Keep the full instrumental and full acapella to hand, already prepared and beatgridded. If the separation artefacts are audible through the venue’s speakers, or that night’s equipment does not even support your usual workflow, perform the mashup using those two rendered layers. The sound may be slightly less flexible, but the routine still works.

---

9. Common mistakes

• Using separation in the booth as the main plan: it consumes resources and makes every set sound different.
• Skipping the trim: a source longer than eight minutes will not fit, and discovering that halfway through the night is too late.
• Improvised names: with individual stems, an ambiguous name is a loading mistake waiting to happen.
• Leaving the beatgrid to chance: stems whose grids have not been checked will not enter in time, even if the separation itself was perfect.
• Forgetting the soundcheck: testing the stems individually does not test the mashup.
• Doing without a backup plan: artefacts that seem subtle on your monitors may be obvious on the venue’s system.