AUDIO SONG FINDER

Identify a song from an audio clip

Upload a short recording or audio file to identify the music playing. Compare possible song titles and recording artists using previews and listening links in your browser.

USE THE TOOL

Upload the clearest part of the song

View an example

An illustrated song search

Static example only. No audio is played or uploaded, no search is run, and no credits are used. This is not a verified recognition result.

1 · Give a clue
Choose a clear 10–15 second clip containing the chorus. This example includes no audio file.
2 · During a real search
After sign-in, the service checks your clue for possible matches. A search can also return no match. This example does not contact the service.
3 · Example result layout
Illustration · not a live result
Bleeding Love
Leona Lewis · 2007

Before choosing a file: trim a clear 10–15 second section containing the chorus or main melody. The duration check happens in your browser; the file is not uploaded until you select Identify song.

Drop your audio hereor click to browse · MP3, WAV, M4A, MP4 · max 5 MB

● Searches require Google sign-in and use 3 credits only when a match is returned.

Who sings this song? Find the title and artist from audio

If you have a recording but do not know the singer, upload a clear 10–15 second section to the audio song finder. It returns possible song titles and artist names with listening links. Compare the actual recording before deciding which match is right.

Can I find a song by an MP3 file?

Yes. Choose an MP3, WAV or M4A file up to 5 MB and sign in with Google to search. A recognisable chorus or instrumental hook is more useful than a clip dominated by speech. Audio is processed for identification and automatically deleted within 24 hours.

Will the artist always be the singer I heard?

Not necessarily. Covers, live recordings and remixes can resemble another version. Check the artist and recording through the listening links. If you only remember words or a melody, switch to lyrics or humming search.

BUILT FOR THE CLUE YOU HAVE

Get a stronger match with the right input

Fingerprint the actual recording

Audio recognition uses the sound itself instead of guessing from a filename or description. A clean chorus or instrumental hook usually gives the strongest signal.

Works with common files

Upload MP3, WAV, M4A, MP4, or WebM files up to 5 MB. Phone recordings, exported clips, and short screen recordings can all be useful inputs.

Verify before you decide

Review the artist, available recording details, previews, and links to listening platforms. Try another section when the first result is uncertain.

HOW IT WORKS

Three simple steps

01
Choose a recognizable section
Trim or select roughly 10–15 seconds where the music is louder than speech, crowd noise, or effects.
02
Upload and identify
Sign in, choose the clip, and run the song finder. The file is sent through an encrypted connection for recognition.
03
Listen to the likely match
Compare the returned preview and streaming links with the section you remember. Save or share the result when it is right.

REFERENCE

What this tool returns, and where it stops

Need a walkthrough first? Read the step-by-step guide: How to identify background music.

Up to five candidate tracks

Each candidate lists title, artist and, where the catalog has it, album and release year, with a confidence figure and listening links.

A note when the clip itself is the problem

Before upload the browser checks duration and signal level. A clip shorter than about 2 seconds, mostly silent, or very quiet is flagged so you can fix it without spending a search.

Repeatable results

Matches are saved to your account history, and a result can be shared by link.

Input, handling and cost

Accepted files: MP3, WAV, M4A, MP4 and WebM
Maximum size: 5 MB per file (enforced by the server)
What is analysed: Audio is converted to mono and loudness-normalised. Clips up to about 16 seconds are matched in full; longer files are tried as up to three 12-second windows (middle, start, end)
File handling: Sent over an encrypted connection to a private storage bucket, passed to the music-recognition provider, and deleted automatically within 24 hours.
Cost: 3 credits, charged only when matches are returned. Searches with no result are refunded automatically.

Known limitations

Recognition matches a specific recording. Live versions, covers, sped-up edits and DJ mixes often fail or return the original.
Speech, crowd noise and room echo over the music lower the match rate far more than a low-quality microphone does.
Very new, unreleased, library or independent tracks may simply not exist in the reference catalog.
A confidence figure is the provider’s score for that candidate, not a guarantee. Always listen to the preview.

Tested examples: what worked, what missed, and what to do

These are 10 real test runs from 9 and 10 October 2026 on public-domain or Creative Commons recordings, 7 that returned the right song and 3 that missed. We chose them so that the mix roughly resembles the share of searches that return a result on the live service, so this is a curated illustration, not a measured accuracy rate. The audio preparation and recognition request match the production service, but the tests were run from a script rather than a user account, so no credits were involved. A different recording, date or catalog can give a different result. Percentages are the confidence shown in the product, and a high figure is not proof that a match is right.

Hit: a famous waltz, clean 15 seconds

Input: 15 seconds from a public-domain recording of Strauss’s Blue Danube.
Result: The Blue Danube by Johann Strauss at 100%. The next candidates were other well-known recordings of the same waltz.
Why it worked: A distinctive, well-documented piece played clearly and without anything covering it.
What to do: Pick the most recognisable passage and keep the recording clean. When several recordings of the same piece appear, use the preview and performer name to find yours.

Hit: right piece, a different performance

Input: 15 seconds from a Creative Commons amateur recording of the second movement of Beethoven’s Moonlight Sonata.
Result: Piano Sonata No. 14, “Moonlight”, II (Wilhelm Kempff) at 96%, followed by other pianists’ recordings. The tested performance itself was not listed.
Why it worked: The piece is in the catalog even though this particular performance is not, so the match identifies the music rather than the exact recording.
What to do: Treat the result as the piece, then check the performer if you need that exact version.

Hit: the exact recording

Input: 15 seconds from a Creative Commons recording of Vivaldi’s Spring, first movement (John Harrison, violin).
Result: “Spring I Allegro”, John Harrison with the Wichita State University Chamber Players, 100%, the recording tested. Other 100% entries were different catalog listings, one credited to Debussy.
Why it worked: A commercially catalogued recording matches exactly, but catalogs also hold duplicate and mislabelled entries.
What to do: Read every candidate’s title, artist and preview rather than only the first, because equal scores can point at different listings.

Hit: a loud orchestral piece

Input: 15 seconds from a public-domain 1921 recording of Wagner’s Ride of the Valkyries.
Result: “Ride of the Valkyries – American Symphony Orchestra 1921” at 100%, a close match to the tested recording.
Why it worked: Strong, distinctive material with a clear signal.
What to do: Choose a section with the full theme rather than a quiet introduction.

Miss: loud noise over the music

Input: The Swan Lake finale excerpt (which matched cleanly on its own) with loud synthetic pink noise mixed in.
Result: “Two-O-Ten” by Cressida at 25%, followed by unrelated tracks. The correct piece was not listed.
Likely cause: The noise level was high relative to the music. The signal the matcher needs was buried, and the low confidence was a fair warning.
What to do: Record close to the speaker, wait for a gap in talking or crowd noise, and re-record rather than boosting the volume afterwards. Don’t rely on a 25% result. Note that in another test a lighter white-noise mix over a very distinctive waltz still matched, so the effect depends on how loud the noise is compared with the music.

Miss: an independent guitar cover

Input: 15 seconds from a Creative Commons solo-guitar recording of Pachelbel’s Canon in D.
Result: One candidate, “Borboletas” by Victor & Leo at 34%, which is not the piece. The correct piece was not returned.
Likely cause: We cannot say for certain, but this performance is probably not in the reference catalog, and the matcher found a different recording with a similar pattern. Covers and independent recordings are the most common cause of this kind of miss.
What to do: A returned result is not always right. Listen to the preview, then try another section of the song, a recording of the original or best-known version, or a different search method such as lyrics or a description.

QUESTIONS, ANSWERED

Frequently asked questions

How long should an audio clip be?

A clear 10–15 second section is usually a useful target. The most recognizable part matters more than overall length.

Can I identify a song from a phone recording?

Yes. Record close to the speaker and reduce nearby speech or background noise whenever possible.

What happens to uploaded audio?

Uploaded audio is kept in private storage for recognition and short-lived troubleshooting, then automatically deleted within 24 hours.