AI Music Genre Classifier

About AI Music Genre Classifier

This tool uses a deep learning AI model trained on over 2 million songs to classify music into 400+ styles across 14 genres. Powered by TensorFlow.js, all processing happens locally in your browser for maximum privacy.

How does AI music genre classification work?

The tool uses a convolutional neural network (CNN) trained on the Discogs music database. It extracts mel-spectrogram features from your audio and compares them against patterns learned from millions of songs to identify the most likely genres and styles.

What genres and styles can be detected?

The AI can identify 400+ music styles across 14 main genres: Blues, Classical, Electronic (House, Techno, Dubstep, etc.), Folk/World/Country, Funk/Soul/R&B (including Contemporary R&B, Neo Soul), Hip Hop (Trap, Boom Bap, etc.), Jazz, Latin (Reggaeton, Salsa, etc.), Pop (K-pop, J-pop, etc.), Reggae, Rock (Metal, Punk, Indie, etc.), and more.

Can this tool detect R&B music?

Yes! Unlike basic classifiers, this AI can accurately detect R&B and related styles including Contemporary R&B, Neo Soul, New Jack Swing, Rhythm & Blues, and Soul. These are categorized under the 'Funk / Soul' parent genre.

How accurate is the AI classification?

The model achieves high accuracy on the Discogs dataset. It performs best on songs with clear genre characteristics. For mixed-genre songs, it shows probability distributions across multiple styles, which is more informative than a single label.

What audio formats are supported?

All common audio formats are supported including MP3, WAV, OGG, AAC, M4A, FLAC, OPUS, and more. Files up to 100MB can be processed. The tool analyzes up to 60 seconds of audio from the middle of the track.

Is my audio data safe?

Yes! All processing happens locally in your browser using TensorFlow.js. Your audio files are never uploaded to any server. The AI model is downloaded once and runs entirely on your device.

Why is the first analysis slower?

The AI model (~50MB) needs to be downloaded and initialized on first use. Subsequent analyses are much faster as the model is cached in memory. The model also 'warms up' with a test prediction to optimize GPU performance.

What's the difference between genre and style?

Genre is the broad category (e.g., Electronic, Hip Hop). Style is the specific subgenre (e.g., Deep House, Trap). The tool shows both: the top detected styles and how they aggregate into parent genres.

Can I see how the genre changes over the track and export the results?

Yes. Each analysis window maps to a real time offset, so the Genre Timeline lists the top genre, style and confidence for every segment - useful for spotting an intro vs a drop vs an outro, tagging cue points or splitting medleys. You can export the full timeline plus the overall top-10 styles as CSV or JSON for spreadsheets, a DAW, a catalog database or DJ/library playlist tagging. Everything is generated locally; nothing is uploaded.

How does the analysis work technically (sample rate, mel bands, window)?

Audio is downmixed to mono, resampled to 16 kHz and turned into a log mel-spectrogram with 96 mel bands (512-sample frames, 256-sample hop), exactly matching the Discogs EffNet input. The model runs on overlapping 128-frame patches; predictions are averaged for the overall result and kept per patch for the timeline. By default up to 60 seconds from the middle of the track are analysed, which captures the main section while keeping inference fast. Treat styles above roughly 20-30% confidence as strong, and use the full distribution for mixed-genre material.