Recipe: rebuild an AI generator's master from its stems in a free DAW

Short answer: load every stem at time zero in a free DAW such as Audacity, pull each track down a few dB so the sum does not clip, then compare it against the full-song download. Expect it to come close but not to match exactly, because split stems are an estimate of the mix, not the tracks it was built from.

What it costs: Audacity is free and open source on Windows, macOS and Linux (Audacity). The stems are the paid part: on Suno, stem splitting is a Pro or Premier feature, and Auto Split uses 50 credits for up to 12 stems (Suno). Stem files do not count as extra downloads beyond the song itself (Suno).

Time: about 30 to 45 minutes for a three-minute song.

The steps

1. Download the full song and the stems as WAV (5 minutes). You need the full song as your reference. If you work in Suno Studio, its audio exports are WAV, and Multitrack exports all tracks as stems "within the context of your Studio mix" (Suno). Keep everything at the same sample rate.

2. Import everything into one project (5 minutes, free). In Audacity, drag the stems and the reference into one project. Make sure every clip starts at 0:00; do not nudge anything yet.

3. Mute the reference and gain-stage the stems (5 minutes, free). The Audacity manual is blunt: "Because mixing tracks is additive, the combined level may show clipping distortion", and its fix is to lower the gain sliders on one or more tracks; peaks that line up reinforce each other (Audacity). Start every stem at about -6 dB and watch the meter on the loudest chorus. For a whole-mix trim, select all tracks and use Amplify with a negative value (Audacity). Our floating-point gain staging guide explains why headroom inside the DAW is cheap but the final file is not.

4. Match levels by ear, then by meter (10 minutes, free). Unmute the reference and switch between it and the stem sum. The reference is a finished master, so it will be louder; lower the reference rather than pushing the stems into clipping. Compare at matched loudness, or the louder one always sounds "better". Our LUFS explainer covers how to read loudness.

5. Run a null test (5 minutes, free). Select the reference and apply Invert, which "flips the audio samples upside-down, reversing their polarity"; mixed with identical audio, the result is cancelled out (silenced) (Audacity). Play the sum against the inverted reference. What you hear is the difference between the two.

6. Read the leftovers (5 minutes, free). A quiet, smeary residue is normal. A loud, clear instrument means a stem is missing or mislabelled; Suno says Auto mode does not always label stems correctly (Suno). A residue that grows louder through the song is timing drift; our guide to fixing generated stems covers locking the tempo before you export.

7. Rebuild your own master (10 minutes onward, free). Once the sum is clean, rebalance, process and export your own version. Treat the generator's master as a reference, not a target. Our home mastering recipe picks up from here.

Why the sum differs from the download

• The stems are separated, not recorded. Suno describes Auto Split as dividing a song into stems (Suno). Our reading: splitting a finished mix is an estimate, so energy that the separator could not assign cleanly either lands in the wrong stem or is lost, and the sum cannot null perfectly. Our stem separation guide covers those artefacts.
• The master bus may not be in the stems. Suno's Studio help says the Full Song export includes all tracks and processing, but does not say whether Multitrack stems carry that processing (Suno). If master compression or limiting is missing from the stems, the sum will be more dynamic and quieter.
• Summing adds level. Even perfect stems summed at unity can clip where the master never did, because peaks reinforce each other (Audacity).
• Phase. Peaks that meet troughs cancel (Audacity), so a stem with flipped polarity or a few milliseconds of offset thins the low end.

What we think

The null test is the most useful ten minutes in this whole workflow, and almost nobody does it with generated stems. It tells you, before you spend an evening mixing, whether the stems are a faithful kit or a rough sketch. We would not expect a perfect null from any split-based stems, and we would not chase one; a sum that sounds like the song at matched loudness is the bar. If it does not reach that bar, fix the source or re-split rather than trying to EQ your way back to the original.

FAQ

Is there a free way to do this? Yes for the DAW: Audacity is free. Getting the stems from Suno needs a paid plan (Suno).

Why is my stem sum quieter than the download? Most likely the stems lack the final limiting the full song had. Do not just raise faders; that is what clips.

Should the null test be silent? Only for stems that were bounced from the same mix. Split stems will always leave some residue.

Sources

• Audacity home page
• Audacity Manual: Mixing
• Audacity Manual: Invert
• Suno: Advanced Stem Separation
• Suno: What types of files can I download?
• Suno: Exporting from Studio

Updated 29 Oct 2026.