ERASYOpen App

AI Music

How to Remove SynthID from Audio (and What It Actually Is)

By Eddie Mathews··9 min read
How to Remove SynthID from Audio (and What It Actually Is) — Erasy

“How do I remove SynthID from my audio?” is one of the most searched questions in AI music — and most of the people asking it do not actually have a SynthID watermark. Before you try to strip something, it helps to know exactly what it is and whether it is even on your file.

Here is what SynthID really is, which tools use it, why Google built it to survive edits, and what the realistic path looks like for a creator trying to distribute AI music.

The short version

SynthID is Google’s audio watermark, used only on its own Lyria model and NotebookLM — not Suno or Udio. Google says it survives noise, MP3 compression and speed changes by design.

If your track is from Suno or Udio, you are dealing with generator fingerprints, not SynthID — and a purpose-built remover like Undetectr is built for exactly that.

What SynthID actually is

SynthID is a watermarking system from Google DeepMind. For audio, Google describes it as embedding “a watermark into any audio generated or published through our AI music generation model Lyria or the podcast generation feature of Notebook LM,” and says it is inaudible to the human ear.

Technically, the waveform is converted into a spectrogram, the watermark is embedded in the frequency domain using psychoacoustic masking — the parts of the sound your ear is least likely to notice — and the audio is reconstructed. The mark is woven into the sound-wave data itself, not added as a separate tone or a metadata tag, which is what makes it hard to remove without a dedicated approach.

Which tools actually use SynthID

Which tools use Google SynthID audio: Google Lyria yes, NotebookLM yes, Suno no, Udio no
SynthID is Google Lyria and NotebookLM only — not Suno or Udio.

This is the part that saves most people a pointless search. SynthID audio is applied to:

SourceUses SynthID audio?
Google Lyria (Google's AI music model)Yes
NotebookLM audio overviews / podcastsYes
SunoNo — its own fingerprint + metadata
UdioNo — content credentials + fingerprint
Stable Audio / RiffusionNo — their own signals

So unless your track came out of Lyria or NotebookLM, it is not carrying SynthID at all. The watermark you are worried about is a different signal entirely.

Why SynthID is built to survive edits

Google SynthID audio watermark is built to be durable and survives MP3 compression, added noise and speed changes
SynthID is engineered to survive common edits.

We are not going to sell you a fantasy here. Google engineered SynthID to be durable, stating the watermark “can’t be altered by common modifications like adding noise, MP3 compression, or changing the speed of the track.” Google also runs a public SynthID Detector that can pinpoint the exact segments of an audio file where the watermark appears.

In other words, the casual tricks people try — re-encoding, adding hiss, nudging the tempo — are exactly what it is designed to shrug off. If you genuinely have Lyria or NotebookLM audio, treat SynthID as a durable, by-design feature rather than something with a quick workaround.

The big misconception

Because SynthID is the most talked-about AI watermark, people assume every AI track has one. They do not. The vast majority of AI music that creators are trying to distribute comes from Suno and Udio, and those carry generator-specific fingerprints and metadata — a related but separate problem with its own, more tractable solution.

What creators actually face

If your goal is to get a Suno or Udio track past distributor detection, the target is the statistical fingerprint those platforms leave, not SynthID. That is a solvable problem: a purpose-built remover analyses what a classifier measures, neutralises those markers, and re-masters the file — leaving the audible music untouched.

The tool we recommend for that is Undetectr, which is built specifically for AI music and handles the fingerprint and the mastering in a single browser pass.

The clean-upload workflow

01
Identify the source

Lyria or NotebookLM means SynthID (durable by design). Suno or Udio means a generator fingerprint you can address.

02
Clean the fingerprint

For Suno/Udio exports, run the file through Undetectr to strip the markers and master to spec.

03
Check for collisions

Verify this yourself against existing releases — no cleanup tool performs the check.

04
Distribute and disclose

Upload to your distributor and disclose AI use honestly where asked.

Our recommended tool

Clean the watermark you actually have

For Suno and Udio tracks, Undetectr strips the fingerprint and masters to spec so the file passes detection. Try it free — no credit card.

Frequently asked questions

What is SynthID for audio?+
SynthID is Google DeepMind's watermarking system. For audio, it embeds an inaudible watermark directly into the sound wave — Google converts the audio to a spectrogram and weaves the mark into frequency ranges the human ear barely notices. It is applied to audio from Google's own Lyria music model and NotebookLM's audio overviews, and is designed to be detectable by Google's SynthID Detector even after edits.
Do Suno and Udio use SynthID?+
No. SynthID audio is specific to Google's Lyria model and NotebookLM. Suno and Udio use their own detection signals — metadata markers, content credentials and spectral fingerprints. So if you are trying to clean a Suno or Udio track, you are not actually dealing with SynthID at all, which is why a 'SynthID remover' search usually points you at the wrong problem.
Can you remove a SynthID watermark from audio?+
Google designed SynthID to be durable and states it survives common changes like adding noise, MP3 compression and speed shifts, so it is deliberately hard to strip. We are not going to pretend otherwise. For the far more common case — a Suno or Udio track with generator fingerprints rather than SynthID — a purpose-built remover like Undetectr targets those statistical markers and re-masters the file so it passes distributor detection.
Is my AI music even watermarked with SynthID?+
Only if it came from Google's Lyria or NotebookLM. If you generated it in Suno, Udio, Stable Audio or Riffusion, it carries that platform's own fingerprint, not SynthID. Identify the source first — it determines what you are actually up against and which cleanup approach applies.