Which AI music detector is the most accurate?

Nobody can answer that honestly today. Ircam Amplify publishes 99% accuracy and AI Voice Detector publishes 99% accuracy, but neither publishes the corpus, the metric definition or the evaluation method behind the number, so the two figures are not comparable to each other or to anything else. Until detectors are measured on a shared public corpus, treat every percentage on every vendor site — including the absence of one here — as a marketing decision rather than a result.

Is there a genuinely free AI song checker?

Yes. This tool is free with no account, no quota and no paid tier, and Deezer's playlist checker is free for listeners. AI Voice Detector has a free tier capped at 50 detections a month. Ircam Amplify is not consumer-priced at all — access goes through a request form.

Which detectors upload my audio?

Ircam Amplify and AI Voice Detector both analyse audio server-side, so your file leaves your machine. Deezer's consumer tool does not take a file upload at all — it reads playlists. This tool decodes and analyses in your browser, so the audio never reaches a server.

What is the difference between an AI music detector and an AI voice detector?

They read different evidence. A music detector looks at spectral behaviour, stereo field, dynamics and transient structure across a full mix. A voice detector looks at the artefacts of synthesised or cloned speech. Running a song through a voice tool, or a voice note through a music tool, produces a confident answer to the wrong question.

Can any of these tools prove a track was AI-generated?

No. Every tool listed here outputs a probability derived from acoustic cues, and all of them can be defeated by re-recording, heavy processing or human post-production on generated stems. A high score is a reason to ask for stems, session files or version history — it is not proof of authorship on its own.

Why compare an AI music tool with an AI voice tool at all?

Because searchers conflate them constantly. People arrive looking for a way to check a song and land on a voice-deepfake product, or the reverse. Naming the boundary is more useful than pretending every tool competes for the same job.

AI music detectors compared, including ours

Four tools that people reach for when they want to know whether audio was generated: this one, Ircam Amplify, Deezer’s detector and AI Voice Detector. What each actually decides, what it costs, where your file goes, and whether its accuracy claim has a published method behind it.

Every third-party claim below is quoted from that vendor’s own public pages, checked 2026-08-15. Nothing here is sponsored or affiliate-linked.

The short answer

These tools are not really competitors. One screens catalogues through an API, one reads playlists against an already-classified library, one detects cloned speech, and this one checks a single music file privately in your browser. The useful question is not which is best but which is answering your question — so the last section says plainly when to use something other than us.

Summary comparison of four AI detection tools
Tool	What it detects	Price	Audio uploaded?	Accuracy claim
AI Music Detector	Generated music	Free	No — runs in your browser	None published
Ircam Amplify	Generated music, at catalogue scale	Enterprise quote	Yes — cloud API	99%, method not published
Deezer	Generated tracks inside playlists	Free	No file upload offered	None, but method peer-reviewed
AI Voice Detector	Cloned and synthetic speech	Free tier, then $49–$599/mo	Yes	99%, method not published

Why two tools both claim 99%

Ircam Amplify advertises “99% accuracy” with “less than 1% false positives”. AI Voice Detector advertises “99% accurate”. Neither figure is published alongside a corpus, a metric definition or an evaluation procedure, and the two tools do not even perform the same task, so the numbers cannot be compared with each other and cannot be reproduced by anyone outside those companies.

An accuracy percentage only means something once you know what was tested and how a correct answer was defined. Detection rate on generated tracks, false-positive rate on human recordings and abstention rate move in opposite directions: any of the three can be made to look excellent by sacrificing the others. That is why this site publishes a benchmark protocol with unmeasured cells instead of a headline number, and why the most transparent entry on this page is Deezer — not because it claims the most, but because its method was peer-reviewed and its code released.

AI Music Detector (this site)

AI Music Detector

What it detects: Whether a music file carries the acoustic signatures typical of fully generated audio. It does not identify which generator made a track and does not analyse speech.
Who it is for: Anyone with a file and a question: listeners, curators, A&R, teachers, moderators, journalists.
How you access it: Free web tool. No account, no email, no API.
Pricing: Free, unlimited, with no paid tier at all.
Where your audio goes: Nothing is uploaded. Decoding and analysis run in your browser with the Web Audio API, and the decoded audio is discarded when you close the tab.
Accuracy claim: None published. We publish a benchmark protocol instead, with result cells marked not measured until an evaluation has been run.
Method published?: Yes — signals, thresholds, ensemble configuration, benchmark protocol and limitations are all public.
Limits: No file-count, size or duration cap beyond what your own browser can decode.
Best for: A private, immediate second opinion on a single file, with reasoning you can read and a PDF you can keep.

Source: Our methodology

Ircam Amplify AI Music Detector

Ircam Amplify, the commercial spin-out of IRCAM in Paris.

What it detects: Fully AI-generated music tracks. The product page names coverage of major generators including Suno, Udio, Sonauto and ElevenLabs.
Who it is for: Music industry professionals — the product page lists DSPs, distributors, labels, CMOs/PROs and publishers.
How you access it: Not self-serve. Access is via a request form, with integration described as a REST API and SDK. A public demo page exists separately.
Pricing: Not published. Enterprise quote only.
Where your audio goes: Server-side by necessity: audio is sent to their cloud API for analysis. Retention terms are not stated on the product page.
Accuracy claim: “99% accuracy” with “less than 1% false positives”, stated on the product page.
Method published?: No corpus, metric definition or evaluation method is published next to that figure.
Limits: Not published for the API. Catalogue-scale throughput is advertised instead.
Best for: Rights-holders and platforms that need to screen a catalogue at scale through an API and can go through procurement.

Source: ircamamplify.io product page

Deezer’s AI music detection

Deezer, working from its own research group.

What it detects: AI-generated tracks in Deezer's catalogue. The consumer-facing tool checks tracks in playlists against that already-classified catalogue rather than analysing a file you supply.
Who it is for: Deezer listeners for the playlist tool; Deezer itself for the catalogue pipeline.
How you access it: Free web tool that scans playlists from a connected streaming service. The catalogue-side classifier is not an external product or API.
Pricing: Free for the playlist tool.
Where your audio goes: No audio upload is offered. The playlist tool matches known tracks, so there is no file for you to hand over.
Accuracy claim: No headline accuracy percentage is published on the consumer pages.
Method published?: Yes, and unusually so: the underlying detector is described in a peer-reviewed ICASSP 2025 paper with code released publicly.
Limits: Playlist-based, so it cannot help with an unreleased file, a private mix or anything not in the catalogue.
Best for: Checking how much of a playlist you already listen to is generated, and reading a genuinely published detection method.

Source: deezer.com AI music detector

AI Voice Detector

AI Voice Detector (aivoicedetector.com).

What it detects: Synthetic and cloned speech in audio and video — voice deepfakes and text-to-speech. It is a voice tool, not a music tool.
Who it is for: Individuals through to enterprises verifying whether a recorded voice is real.
How you access it: Web app plus a Chrome extension. File upload, URL, or record in the page.
Pricing: Free tier at $0 with 50 detections a month, then paid plans published at $49, $199 and $599 a month.
Where your audio goes: Files are uploaded to their service for analysis.
Accuracy claim: “99% accurate”, stated on the homepage.
Method published?: A dated methodology page is referenced, but no corpus or metric definition backing the 99% figure was found on the public pages.
Limits: The free web tool states a 25 MB file cap, and the free plan caps each detection at two minutes.
Best for: A suspected cloned voice in a call recording, voice note, podcast or video — the case this site deliberately does not cover.

Source: aivoicedetector.com pricing

Which one you should actually use

You have one file and want an answer now, without handing it over: Use this tool. Nothing is uploaded, there is no account, and you get the reasoning behind the score rather than a bare verdict.
You need to screen thousands of tracks automatically: Use a commercial API such as Ircam Amplify. A browser tool cannot batch a catalogue, and we do not offer an API.
You think a speaking voice was cloned: Use a voice-deepfake tool such as AI Voice Detector. Music detection reads production and spectral behaviour, which is the wrong evidence for speech.
You want to know how much AI is in what you already listen to: Use Deezer's playlist checker. It works from a classified catalogue, which is a question no file-based detector can answer.
You need something defensible for a dispute or a payout decision: No tool on this page, ours included, is evidence. Every one of them produces a probability from acoustic cues, and none can prove authorship. Treat any result as a reason to ask for the session files, stems or project history.

Corrections

Vendor pages change, and a comparison that goes stale becomes misinformation. If something here is out of date or wrong about your product, tell us through the contact page and we will correct it and re-date the page. See also our accuracy and limitations, the methodology, and the benchmark protocol.

Frequently asked questions

Nobody can answer that honestly today. Ircam Amplify publishes 99% accuracy and AI Voice Detector publishes 99% accuracy, but neither publishes the corpus, the metric definition or the evaluation method behind the number, so the two figures are not comparable to each other or to anything else. Until detectors are measured on a shared public corpus, treat every percentage on every vendor site — including the absence of one here — as a marketing decision rather than a result.

Want to check a file right now?

Free, no account, and your audio never leaves your browser.

Open the detector