Song FinderHow to Identify a Song: Audio, Lyrics, a Video Clip or a Vague Memory
Six ways to name a song you cannot name — by audio, by a line of lyrics, from a video link, by humming, by describing the mood. Which works for what, and what to do when recognition fails.
What you have determines what works. Start from the right one and this takes seconds; start from the wrong one and you can spend an evening on it.
| What you have | Start here |
|---|---|
| An audio file or clip | Audio recognition |
| Music playing near you | Record it |
| A line you remember | Lyric search |
| A link to a video or reel | Paste the URL |
| Only a melody in your head | Humming search |
| Only a feeling | Describe the mood |
1. You have the audio
This is the easy case. Acoustic fingerprinting turns a few seconds of audio into a compact signature and matches it against a database of millions.
Upload a file, paste a media URL, or record from your microphone at the song recognizer. Your clip runs through multiple independent engines in sequence — each with its own fingerprint database — so a miss on the first does not end the search.
How long a clip? Ten to fifteen seconds of the clearest music you have. Longer does not help; cleaner does.
Pick the right part. Intros, breakdowns and fade-outs sound alike across thousands of tracks. The chorus, or whatever is most distinctive, fingerprints far better. Use the waveform selector to grab exactly that section.
2. Music is playing near you
Same method, live input. Hold the microphone as close to the source as you can and record.
Recognition drops off fast with crowd noise, room echo and distance from the speaker. If it fails, the fixes in order:
- Get closer. More useful than anything else on this list.
- Record a different section. Wait for the chorus.
- Clean it up. Run the recording through noise reduction for crowd and air conditioning, or normalize if it came out too quiet, then try again.
3. You remember some of the words
Fingerprinting is useless here — there is no audio. Text search is the tool.
Search the line you remember with the lyric finder.
What actually works:
- Four to eight words is the sweet spot. Two words match everything; a whole verse matches nothing because you have misremembered part of it.
- Quote an exact phrase if you are confident:
"we were both young when I first saw you". - Pick the odd words. Unusual nouns beat common verbs. "Hallelujah" narrows; "baby" does not.
- Skip the chorus if it is generic. A strange line from a verse is worth more than a hook everyone has used.
When you are misremembering — and people usually are — drop to the two or three words you would swear to and let the rest go.
4. You have a link to a video or reel
Paste the URL directly into the recognizer. The audio is pulled from the link and fingerprinted the same way as a file.
The catch with short-form video: TikTok, Reels and Shorts routinely play music sped up or slowed down. Pitch-shifted audio has a different fingerprint and can fail to match at all.
If it does, use Speed & Pitch to shift it back toward normal and run recognition again. Sped-up edits are usually somewhere between 110% and 130% of the original — try 0.85× and 0.8× and re-run.
5. You only have the melody
Fingerprinting needs the actual recording, so humming will not match it. Melody search is a different technology, and Google's "hum to search" or SoundHound are the ones to use.
Hum steadily for ten seconds or more, and hum the most distinctive part rather than the opening.
Once you have a title, come back and look up its BPM and key if you need the numbers.
6. You do not even have that
Sometimes what is left is "something like the song from that film, sad, piano, a woman singing."
Two things to try.
Describe the feeling. The mood finder searches on scene and mood — late night drive, rainy Sunday, gym at 6am — rather than title or lyric.
Work from a song you do know. If you can name anything that sounds close, Similar Songs Finder will give you tracks near it in key, tempo and feel. The one you are chasing is often two or three steps away.
When recognition keeps failing
Work through these in order. Change one thing at a time, and re-run recognition after each — stacking three fixes at once means you learn nothing about which one helped.
| Symptom | Fix |
|---|---|
| Crowd noise, room echo | Noise reduction, then retry |
| Recording is very quiet | Normalize, then retry |
| Sped-up or slowed edit | Speed & Pitch back toward 1.0× |
| Speech over the music | Select a music-only region with the waveform tool |
| Only the intro was captured | Select the chorus instead |
| It is a live or cover version | The fingerprint differs from the studio master — try lyric search instead |
The case that never works: a fingerprint only matches the specific recording it came from. A live take, a cover, a remix or an AI-generated track will not match the studio original no matter how clean your audio is. Switch to lyric search — the words usually survive when the recording does not.
Once you have the title
- Song BPM & Key Finder — tempo, key and Camelot code
- Music Genre Finder — genre and the finer style tags behind it
- Similar Songs Finder — more like it
Frequently asked questions
How do I identify a song without an app? Everything here runs in a browser with no install and no account — upload audio, paste a link, record from the microphone, or search a lyric.
Can I find a song by humming it? Not with fingerprinting, which needs the original recording. Melody search — Google's hum-to-search or SoundHound — is built for that.
How many seconds of audio does recognition need? Ten to fifteen seconds of clear music. A short clean clip beats a long noisy one.
Why can't it identify a live version? Acoustic fingerprints identify a specific recording. A live performance is a different recording and produces a different fingerprint. Search the lyrics instead.
Can I identify music from a TikTok or Instagram video? Yes — paste the link. If it fails, the audio has probably been sped up; shift it back with Speed & Pitch and try again.
Is my audio uploaded anywhere? The audio tools — noise reduction, normalize, speed and pitch, trimming — run entirely in your browser and the file never leaves your device. Recognition itself needs to send the clip to a fingerprint service, since the database is what does the matching.