You want to mix a Suno song properly, and the vocal is where it shows first. Before a compressor or an EQ goes on it, the vocal should be on its own track, as a lossless file, without other instruments in the gaps and without a metallic layer on the voice. This guide walks through that in order, starting with the export in Suno, and marks the two steps where our plug-ins fit.
We make Intrect de-leak and Intrect de-artifact, named in steps 5 and 6. Both are optional, we state what they do not do, and we name other tools for each step.
The short answer
Export the vocal as its own WAV stem in Suno, then check it before you process it. In Suno, use Get Stems with Split from Mix to pull out the vocal plus a complement track, and download WAV. Import it next to the full mix, solo it and listen to the gaps. Reduce bleed first (Intrect de-leak or a re-split), then metallic or watery residue (Intrect de-artifact), then de-ess and start your normal vocal chain. Keep the original stem untouched.
1. Export the vocal stem in Suno
Suno’s help centre (as of November 16, 2026) describes Get Stems as a Pro and Premier feature. In your Library or Workspace, open the More Actions (…) menu next to the song, choose Get Stems, pick a split type and select Extract. In Suno Studio, right-click the clip on the timeline and choose Split Stems.
| Split type | What you get | Plans | Cost |
|---|---|---|---|
| Split from Mix | One instrument or vocal, plus a complement track | Pro, Premier | 10 credits per stem |
| Auto Split | Up to 12 stems | Pro, Premier | 50 credits per extraction |
| Advanced Split | Stems you choose from nearly 100 instruments | Premier only | 10 credits per stem |
If you only need the vocal, Split from Mix gives you the vocal and everything else as two files, which is enough to mix a vocal against its backing. Take Auto Split if you also want to rebalance the instruments; our guide to cleaning all Suno stems covers that case.
Download WAV, not MP3. Suno’s file-type page lists WAV for Pro and Premier and MP3 on all plans, and says stems do not count as downloads beyond the song itself. An MP3 has already discarded part of the signal, and converting it to WAV later does not bring it back. Download the full mix too, as a reference.
No stems on your plan? Separate the vocal yourself from the best mix file you have, with an open-source separator such as Demucs or a separation service. Everything below still applies, and bleed is more likely.
2. Import it so you can compare
- Put the vocal stem, the complement track and the full mix in one session, all at the same start point.
- Duplicate the vocal and work on the copy. Mute the original and keep it.
- Set the session tempo before you edit, and check vocal transients against the grid before you quantise anything.
- Play vocal plus complement against the full mix at matched loudness. If the vocal jumps forward, drops out or flams, fix that first. Do not expect the two to cancel exactly: Suno does not say its stems add back up to the mix sample for sample, so treat this as a check for differences, not a null test.
3. Gain-stage before you judge
Trim the vocal so it peaks well below 0 dBFS and sits at a normal level against the complement. Louder sounds better, so level-match before you decide it needs processing.
4. Listen for the two problems
Loop a phrase with a held note and a gap, and solo the vocal.
| What you hear | What it is | Step |
|---|---|---|
| A snare, chord or backing part between phrases | Bleed: another part left in the vocal stem | 5 |
| Silent gaps, but a metallic, glassy or watery layer on the voice | Residue: generation or separation damage on the voice | 6 |
| Clean gaps and a clean voice | Nothing to clean | Skip to 7 |
One stem can have both. We have not measured Suno’s current built-in splitter, so we make no claim about how clean your stem will be. Listen before you process.
5. Reduce bleed first
Where Intrect de-leak fits. de-leak reduces instrument bleed in an already separated vocal stem. Its Gate works in 16 bands, Leak Clean uses a separate 64-band model and can run on the vocal alone, and Residual lets you hear what was removed. If Residual contains the singer, reduce the processing. It runs as a plug-in in your DAW with a 14-day trial; see the de-leak product page for formats.
It does not separate a full mix into stems, recover anything the splitter removed, or suit live monitoring. Other options: re-split the song with another split type, SpectraLayers De-Bleed, or clip gain and spectral edits when there are only a few obvious hits.
6. Then reduce metallic or watery residue
Where Intrect de-artifact fits. de-artifact reduces metallic shimmer and watery textures in generated vocals and separated stems. Insert it on the vocal and load the Vocal preset, made for AI-separated vocal stems. Switch on Residual: you should hear fizz and ringing. If breaths and consonants disappear into it, lower Strength. A/B against Bypass at matched loudness. It has a 14-day full-feature trial with no card, and try.intrect.io plays a free 30-second preview without an account.
It does not fix wrong words, pitch or timing, and the removed signal can include detail you want, which is what the Residual check is for. Our guide to fixing AI music artifacts in Suno vocals lists each symptom and what to use when it is not residue.
7. Now the normal vocal chain
- High-pass below the voice.
- De-ess the splashy s and t sounds.
- Corrective EQ, and a dynamic cut for any narrow ringing band.
- Compression for level, then automation for the words that still jump.
- Reverb and delay last, on sends, so the cleaned dry vocal stays editable.
Bounce the cleaned vocal to a new WAV at the session sample rate before you move on, so the mix does not depend on the cleanup plug-ins staying loaded.
Checklist
- Vocal stem exported with Split from Mix (or Auto Split), downloaded as WAV.
- Full mix downloaded as a reference.
- Original stem kept; working on a copy.
- Aligned and compared against the full mix.
- Gaps checked: bleed reduced first, then residue.
- Residual of every cleanup tool checked for the singer.
- Normal vocal chain after cleanup; reverb last.
Found bleed in a stem?
de-leak has a 14-day trial for macOS and Windows hosts that load VST3 or CLAP. Comparing tools first? Read the bleed comparison.
Related guides
- How to fix AI music artifacts in Suno vocals
- Why Suno stems sound bad, and how to clean them
- How to clean a separated vocal stem
- Remove instrument bleed from a vocal stem
- How to remove Suno shimmer from vocals
- AI music restoration: what to run first
Sources
- Suno Help Center, How do I get stems?, help.suno.com/en/articles/13925185, read November 16, 2026.
- Suno Help Center, What types of files can I download?, help.suno.com/en/articles/13926081, read November 16, 2026.
- Our de-leak and de-artifact product pages and the de-artifact manual, read November 16, 2026.
FAQ
How to clean Suno vocals before mixing?
Export the vocal as its own WAV stem with Get Stems and Split from Mix, import it next to the full mix, and solo it. Reduce bleed first (Intrect de-leak or a re-split), then metallic or watery residue (Intrect de-artifact), then de-ess and start your normal vocal chain. Keep the original stem untouched.
Which Suno split type should I use for the vocal?
If you only need the vocal, Split from Mix extracts the vocal plus a complement track for 10 credits per stem on Pro and Premier, according to Suno's help centre. Auto Split gives up to 12 stems for 50 credits if you also want to rebalance the instruments.
Do I need a plug-in to clean a Suno vocal?
Not always. If the gaps between phrases are silent and the voice has no metallic or watery layer, go straight to your normal vocal chain. Reach for a bleed remover or an artifact remover only for the problem you can hear.