CoversBR: Metadata and Features for Brazilian Cover Song Identification

CoversBR is a large audio database focused predominantly on Brazilian music for cover and live song identification tasks. It comprises metadata and extracted features from 102,298 songs, organized into 26,366 cover groups, totaling approximately 7,070 hours of audio. The dataset is provided by Dirceu G Silva via AWS Open Data, but the original audio files are not included due to copyright restrictions.

Use Cases

• Training models for cover song identification based on metadata and extracted audio features.
• Developing systems for live song identification using audio streamings from radio and TV.
• Analyzing musical similarity and version relationships within the 26,366 cover groups.
• Benchmarking audio feature extraction techniques on a large-scale Brazilian music collection.

Strengths

• Contains metadata and features for 102,298 songs, a substantial collection for research.
• Organizes songs into 26,366 distinct cover groups, providing a structured basis for similarity tasks.
• Total audio duration is approximately 7,070 hours, with an average song length of 240 seconds.

Limitations

• Original audio files are not available due to copyright restrictions, limiting direct audio analysis.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

Dirceu G Silva via AWS Open Data.

Collection Method

Audio streamings captured from radio and TV channels, with features extracted from songs.

Time Range

null

Freshness

Last update date is unknown; freshness unverified.

Geography

Primarily Brazilian music.