GTZAN Audio Collection with 10 Music Genres

Containing 1000 audio tracks, each 30 seconds long. It includes 10 music genres, with 100 tracks per genre, and provides both raw WAV files and 8000 derived Mel Spectrograms. The audio files are 22050Hz Mono 16-bit WAVs.

Use Cases

• Train a genre classifier using the 10 genre labels on the 1000 audio tracks.
• Extract features from the 30-second WAV files for music information retrieval tasks.
• Analyze Mel Spectrograms from 2-second audio windows for temporal pattern recognition.
• Benchmark audio classification models on the balanced set of 100 tracks per genre.

Strengths

• Contains 1000 audio tracks, providing a substantial sample size for analysis.
• Features a balanced representation with 100 tracks for each of the 10 music genres.
• Includes 8000 Mel Spectrograms derived from 2-second audio windows, enabling spectral analysis.

Limitations

• Each audio track is only 30 seconds long, which may not capture full song structures.
• The dataset's age and original collection method may introduce unknown biases in genre representation.
• The specific recording sources and artist demographics for the tracks are not detailed.

Provenance

Source

Hugging Face, uploaded by Lehrig.

Collection Method

Collection of 1000 pre-existing audio tracks, with subsequent processing to create Mel Spectrograms.

Freshness

Last updated on 2022-06-13.