Open SongFinder

What SongFinder cannot do

Checked against the running server on 2026-10-28

Every tool has an edge. These are ours, stated as numbers rather than adjectives, so you can decide in ten seconds whether this tool is the right one for the job.

The short version

  • Refused outright. A YouTube link, a Douyin link, a file on your phone, a hummed melody, a song download. Four of those never reach the matching step at all, so retrying them is pointless.
  • Accepted. A public TikTok link, a public Instagram Reel, Reels or post link, or the direct address of an audio or video file. Anything else is a different tool.
  • Metered. Twenty attempts per address per hour, and one shared pool of a thousand per day for the whole site. Both are hard stops, not warnings.
  • Not guaranteed. An identification can come back empty on bad audio. That is the expected behaviour of audio fingerprinting, not a bug report.
The daily ceiling is the one people misread. One thousand a day is a single counter for the entire site, shared by every visitor at the same moment — not a thousand each. If the tool suddenly says the daily limit is reached while you have barely used it, someone else already spent the pool.

The twelve limits, with the numbers

1. A YouTube link is refused

YouTube is not on the accepted list. A watch page, a short link and a share link all come back as an unsupported platform before any audio is fetched. Only TikTok links, Instagram Reel, Reels or post links, and direct audio or video file addresses are read. A link that is not TikTok or Instagram is fetched as a direct media URL, and what comes back decides the answer: a web page comes back as UNSUPPORTED_PLATFORM before any audio is analysed, so a proxy page that merely wraps a YouTube video is not a way around it.

2. You cannot upload a file from your device

There is no file picker and no form that accepts binary uploads. The tool takes a public address and fetches the media itself. If the audio only exists on your phone, this tool cannot reach it.

3. You cannot hum, sing or use the microphone

There is no microphone input and no recording step. Humming recognition needs you to reproduce the melody well enough for a matcher in a quiet room. This tool works from a link instead, which is a different problem solved a different way.

4. You cannot download the song

A successful identification returns a title, an artist and links to listen on a music service. No audio file is written back to you, and no download link is generated. Public does not mean licensed, and a video being public is not permission to reuse its music.

5. One attempt tries up to four 15-second windows

The audio is cut into 15-second windows, converted to mono, and sent for matching. Without a start time the server works backwards from the end of the media and tries at most four windows — the last 15 seconds first, then 75, 50 and 25 percent of the usable length — stopping at the first match. A start time pins a single window. Each window is 15 seconds because a short, music-rich sample matches better than a long one full of speech and noise.

6. Files up to 100 MB

The fetch step reads the declared content length and refuses anything above 100 MB. The check happens before the body is downloaded, so an oversized file is rejected quickly instead of being pulled down and then discarded.

7. Media up to 10 minutes

A media duration above 600 seconds is refused. The probe reads the duration first, so a long file is turned away before the extraction step runs. When the duration cannot be determined at all, the attempt continues rather than being blocked on a guess.

8. Twenty identifications per address per hour

A twenty-first attempt from the same address inside the same hour is answered with HTTP 429 and a retry delay. The response tells you how many minutes remain, so the page can show a countdown instead of a dead end. This limit is per client address, not per person, so several people behind one shared connection share the twenty.

9. One thousand identifications per day, shared by everyone

The daily ceiling is not per visitor. It is one counter for the whole site, and it is shared by every visitor at once. When it is reached, a new attempt is told the daily limit has been reached rather than failing silently. It resets at midnight UTC. There is also a monthly ceiling well above the daily one, so the daily figure is what you will meet first in practice.

10. Two identifications run at once, up to ten more wait

Two run concurrently. A further request waits in a queue of at most ten. One that gets no slot within 30 seconds is turned away with the same rate-limit answer, and a queue that is already ten deep refuses the next request immediately instead of making it wait. Under heavy load a request can be refused even when your own hourly allowance is untouched.

11. One identification gives up after 90 seconds

The matching step has a hard 90-second ceiling. Fetching the media, cutting the snippet and asking the recognition service all have to finish inside it. A very slow origin server therefore ends as a timeout, not as a match or a miss. Ninety seconds is also the real number to quote: the queue wait is separate and shorter.

12. A missing match is a normal result, not an error

If the 15 seconds you captured are mostly speech, applause, wind or a mix, there is nothing for the database to fingerprint and no title comes back. A file with no audio track ends the same way. Very new, very niche, live and cover versions are also often absent. None of these are failures of the link you pasted, and retrying the same link usually produces the same answer. Moving the start time to a cleaner stretch is the one change worth making.

Which answers mean what

Four outcomes are worth telling apart, because only one of them is worth retrying.

  • A title comes back. The snippet matched. Nothing else to do.
  • No match. The audio was read and nothing in it was recognised. Retrying the same link gives the same answer; moving the start time to a cleaner stretch sometimes does not.
  • No audio. The address resolved but the media has no audio track worth cutting. A different link is the only fix.
  • Unsupported platform. The address was never fetchable by this tool. No amount of retrying changes that.
One trap for anything reading the endpoint directly: a refusal is delivered inside a normal HTTP 200 response, as an error code in the body, not as an error status. A client that checks only the status code will read a refusal as a success and then find nothing in the payload. The rate limit is the exception — that one does come back as HTTP 429 with a retry delay.

What is not stored

The snippet exists only long enough to produce an answer and is deleted when the answer is returned. There is no account, so there is no history to log into, no saved list and no profile. Nothing about a visit is retained as a listening record, because nothing is tied to a person in the first place.

What this tool is not

Not affiliated with any platform

SongFinder is independent. It is not affiliated with, sponsored by or endorsed by TikTok, Instagram, YouTube or any music service, and a refusal or a match says nothing about what those platforms would do.

Not a scientific instrument

Audio fingerprinting is a pattern-matching technique, not a measurement. It is not validated for any evidentiary, legal or commercial use, and a match is a best guess with a confidence behind it rather than a certified fact.

Not the recogniser itself

The matching is done by a third-party music-recognition service. SongFinder fetches the audio, cuts the snippet and passes it on. Knowing that matters when you are deciding what to do with the answer, because the accuracy ceiling belongs to that service, not to this page.

SongFinder

Paste a public TikTok, Instagram Reels or direct media link. The server listens to a 15-second snippet and names the song.

Open SongFinder

More from SongFinder

What song is this? Identify it from a video link Find the song in a TikTok video from its link Find the song in an Instagram Reel