Why Shazam Can't Find a Song: How Music Recognition Works

How Shazam Turns a Clip's Sound Into a Digital Fingerprint

If Shazam or a similar app fails to identify a track that is clearly playing in a clip, the cause is not necessarily that the app "didn't hear" the melody: the recording may simply be missing from the service's catalog, the audio in the clip may have been altered — sped up, slowed down, or mixed with another track — or the sample itself may be too short or too noisy. According to the algorithm's creator, Avery Wang, and Apple's official documentation, Shazam is not searching for a melody as such but for a match between the audio's digital fingerprint and a specific recording already entered in the service's catalog. The fingerprint is built as a "constellation" of spectrogram peaks from the recording, plus a set of hashes computed from pairs of those points — producing a compact code unique to that particular phonogram. From there the search runs not on meaning or on a melody a human can recognize, but on an exact match of those hashes to a recording in the database.

Apple describes the same principle more simply: the app builds a unique digital audio fingerprint and compares it against a catalog of millions of tracks. Developers have a separate tool, ShazamKit, which can match audio not only against Shazam's general catalog but also against a developer's own audio database — for example, the tracks from a specific video or podcast; according to Apple, the audio itself is never sent to Apple, and audio signatures cannot be reversed back into sound. Shazam belongs to Apple: per the company's official press release, the deal closed on September 24, 2018.

Live Performances and Studio Recordings Are Different Objects to the Algorithm

In his 2003 paper, Wang states directly that the algorithm is aimed at recognizing files already present in the database, and is not designed to extend that result to live recordings of performances. In practice this means: if the catalog holds a studio version of a track, but what's being recognized is a recording of a live concert performance of the same song, a match may not turn up — because to the algorithm these are two different objects, even though a person recognizes the melody without hesitation. That is not the same thing as "performs poorly with audio picked up by a phone microphone": according to the paper itself, the algorithm was designed from the outset for exactly that scenario — sound captured by a phone's microphone against background noise, not studio silence.

At the same time, per Wang's description, the system is highly sensitive to exactly which version of a track was played: according to the algorithm's creator in the 2003 paper, it can tell one performance apart from another even when the difference is inaudible to the human ear — meaning it can avoid confusing several similar recordings of the same song by the same artist. The same paper also notes false positives in the system: according to the author, it often turned out that the algorithm had not actually erred but had flagged a genuine case of sampling or melodic borrowing. So the sources give no grounds to claim that confusing versions is impossible in principle — only that distinguishing versions is built into the algorithm's own logic.

The Path Into Shazam's Catalog: Approved Distributors and Editorial Exceptions

Before looking at why a remix or a homemade edit behaves differently, it helps to understand how a track gets into Shazam's catalog at all. Per Apple's official guide, for music to appear in Shazam and Apple Music, the rights holder needs to work with an Apple-approved distributor — meaning the recording has to go through the standard content-supply channel, not simply be played somewhere.

The same Apple guide states directly that the company may hide some content from the catalog for editorial reasons, including misleading content and copyright infringement. Among the examples Apple itself cites are single-file DJ mixes, megamixes, audio fragments, and edited viral clips from TikTok, SoundCloud, Instagram, and YouTube. In other words, even going through a distributor does not guarantee that a particular version stays visible in the catalog.

The Difference Between an Official Remix and a Homemade Sped-Up Edit

This points to a practical line worth keeping in mind before calling Shazam "broken." If a remix, cover, or sped-up version is officially released as a separate track through an approved distributor, it can in principle get its own catalog entry and be recognized — but under its own title, not as a stand-in for the original. A homemade version — a track sped up or slowed down solely for one specific clip and never released anywhere — works differently: it doesn't go through a distributor and doesn't get a separate catalog entry.

Technically, changing speed and pitch shifts the position of the peaks on the spectrogram, and with them the hashes built from those peaks — so it's reasonable to expect that matching such a version against the original recording's hashes becomes less likely. But that is an inference from the described mechanism, not a separately measured result: Wang's 2003 paper tests the algorithm's robustness to noise and compression, not to changes in tempo or pitch, and none of the sources used for this piece give a numeric figure, for today's versions of Shazam, Google, or SoundHound, on how often such a homemade version still turns out to be recognized or unrecognized.

The Algorithm Creator's Early Experiment: Sample Length and Noise Level

This topic has an older, more measurable study behind it — though not about remixes, but about sample length and background noise. In a controlled experiment the algorithm's author ran on a test database of roughly 10,000 tracks, recognition predictably dropped as the signal-to-noise ratio fell and the sample length shortened: the 50% recognition threshold for uncompressed 15-, 10-, and 5-second samples was reached at around −9, −6, and −3 dB SNR respectively, and after GSM codec compression, at around −3, 0, and +4 dB.

That is a 2003 study of one specific early version of the algorithm, not a test of the 2026 app, and the exact figures should not be carried over to today's Shazam. They're cited here only as the source of the underlying pattern: the shorter and noisier a sample, the lower — by the logic of how the algorithm is built — its chance of matching. That is a property of the algorithm as described in 2003; whether today's version is built and behaves the same way is something Apple has not published, and the specific decibel thresholds for it are unknown.

Automatic Recognition and Offline Mode Inside the Shazam App

Besides manual launch, Shazam has several built-in ways to catch music without pressing the button again. The Auto Shazam feature can keep trying to recognize the music around the user without waiting for a manual launch: per Apple's current guide, on iPhone, iPad, and Apple Vision Pro it's turned on with a long press on the Shazam button inside the app, and the Android version matches sound against Shazam's database even while the user has switched to another app. Separate from Auto Shazam, iPhone and iPad offer quick access to a one-time recognition through Control Center — the same place as on Mac and Apple Watch, just via a "Recognize Music" button.

The offline behavior Apple describes applies to ordinary use of the Shazam app, not only to Auto Shazam: if there was no internet connection while recording, the app still builds the audio's digital fingerprint and offers a match later, once a network becomes available — this is confirmed for both the iPhone/iPad and the Android version. A separate caveat concerns headphones: per Apple's guide, Shazam on iPhone and iPad cannot identify a track playing through headphones or a headset if there is no internet connection at that moment — offline mode does not help in this scenario.

A Different Search Principle: How Google Finds a Song From a Hummed Melody

If Shazam didn't find the track, that doesn't mean no tool at all will find it. Google has a separate "Search a song" feature in its mobile app: per Google's own instructions, it lets a user either play a recording or hum, whistle, or sing the melody — offering a second search method that isn't tied to an exact match with the original phonogram.

According to Google's engineers, the humming-recognition mode doesn't build a fingerprint of a specific audio recording; instead, it builds a compact digital representation of the melody from the spectrogram and matches it directly against original recordings, with no intermediate "hummed" reference stored in the database. In the 2020 write-up accompanying this feature's launch, Google states that training recordings with randomly altered pitch and tempo were used so the model would learn to ignore such differences the same way it ignores background noise — even though two performances of the same melody can, in practice, also differ in key. The feature was built on Google's earlier systems for recognizing music that is already playing. In other words, Shazam's failure to match a fingerprint against its catalog doesn't mean the melody can't be recognized by some other method in principle.

SoundHound: A Separate Recognition Service With Its Own Feature List

SoundHound Inc. offers a separate app, SoundHound, which — per its official App Store description and its product page — accepts both a recording playing nearby and a melody the user sings or hums by voice.

From the sources examined for this piece — the app's official page and its App Store listing — only the list of supported audio-input methods is known: playing a recording, or singing or humming by voice. These sources say nothing about the technical side of how SoundHound matches audio.

Practical Steps When the App Fails to Identify a Clip

When the answer is needed not for general understanding but right now, a reasonable order of steps looks like this. If the clip is being watched on the same device where Shazam is installed, and there's an internet connection at that moment, it's worth first trying recognition right on the device: on iPhone and iPad, Shazam can identify music the user is listening to through headphones or a headset, and on Android there's an on-device sound-recognition mode — with or without headphones — plus the Pop-Up Shazam feature, which, per Apple, identifies music playing within the same app the person is using. Another option on Android is to run a song search through Circle to Search, if it's available on the device.

If that doesn't work, or the music is playing from somewhere nearby rather than from the phone itself, it's worth trying to hum or whistle the same tune through Google's "Search a song" feature — a different search principle, not tied to an exact fingerprint of the original phonogram. It's also worth opening SoundHound and trying again the same way: singing or humming the melody by voice, or playing the recording near the microphone.

If there's a reason to suspect an official remix or sped-up version is playing, it's worth searching for the track by the clip's title or creator on streaming services — such a version might be catalogued under its own title, different from the expected original. And only if the recording was made through the device's built-in microphone (not headphones) and there was no internet connection at the time, is it worth simply waiting for a connection: Shazam on iPhone, iPad, and Android will offer a match later, once a network appears, provided the fingerprint was already built and the track is in the catalog. Through headphones or a headset, offline mode won't help at all — that scenario, per Apple's guide, requires an active network connection at the moment of listening. None of the sources this piece relies on describes a universal method that would work for every clip under every condition.