Audio Fingerprint Tool

Explore how Shazam-style audio fingerprinting works. Record short audio clips, visualize their spectral peaks as constellation maps, generate fingerprint hashes, and compare two recordings for similarity — all processed locally in your browser.

Audio Fingerprint Tool

Understanding Your Results

Fingerprint Hash

The hash display shows the computed fingerprint as a series of hexadecimal values. Each hash encodes a pair of spectral peaks: their frequencies and the time delta between them. Identical or very similar audio produces matching hashes, which is the foundation of audio recognition.

Constellation Map

The constellation map plots detected spectral peaks as dots on a time-vs-frequency chart, with lines connecting nearby peak pairs. This sparse representation is what makes fingerprinting robust — it captures the most distinctive features of the audio while ignoring noise and volume differences.

Similarity Score

When two recordings are fingerprinted, the tool computes a similarity percentage based on how many hash values match between them. 90–100% indicates the same audio content, 50–89% suggests similar but not identical audio, and below 50% means the recordings are substantially different.

Peak Density

The peak density slider controls how many spectral peaks are detected. A lower density produces fewer, more distinctive peaks — faster matching but less detail. A higher density captures more peaks for better accuracy at the cost of more data.

How Audio Fingerprinting Works

Audio fingerprinting is the technology behind services like Shazam and SoundHound that can identify songs from short recordings, even in noisy environments. The core insight is that a song's identity can be captured by its most prominent spectral features — the specific frequencies that peak at specific moments in time.

Spectral Peak Extraction

The process begins with a short-time Fourier transform (STFT), which divides the audio into overlapping windows (typically 1024–4096 samples) and computes the frequency spectrum of each window. This produces a spectrogram — a 2D map of frequency magnitude over time. From this spectrogram, the algorithm identifies local maxima: points that are louder than their neighbors in both time and frequency. These peaks represent the most energy-dense moments in the audio and are remarkably stable across different recording conditions, volumes, and background noise levels.

Constellation Maps and Combinatorial Hashing

The detected peaks form a constellation map — a sparse scatter plot of (time, frequency) coordinates. To create searchable fingerprints, the algorithm pairs each peak with several nearby peaks within a target zone. Each pair is encoded as a compact hash: hash = freq1 + freq2 + time_delta. This combinatorial hashing approach generates hundreds of hashes per second of audio, creating a robust fingerprint that can tolerate minor variations. The hash values are independent of absolute time position, making them suitable for matching against a database regardless of where in the song the recording started.

Matching and Hamming Distance

To compare two fingerprints, the algorithm counts how many hashes they share. For binary hash representations, the Hamming distance measures the number of differing bits between two hash values. A low Hamming distance between hash sets indicates high similarity. In production systems like Shazam, matching also considers the time coherency of matching hashes — not just that two recordings share the same hashes, but that those hashes occur at consistent time offsets relative to each other. This dramatically reduces false positives and enables identification from recordings as short as 3–5 seconds.