Memo2496: 2,496 Instrumental Songs with Expert Valence-Arousal Annotations

Memo2496 is a music emotion recognition dataset containing 2,496 instrumental songs annotated by 30 music experts. The dataset provides valence-arousal labels and extracted acoustic features to support affective computing research. It was updated by Qilin Li in April 2026 and is hosted on figshare under a CC-BY-4.0 license.

Use Cases

• Training music emotion recognition models based on expert-annotated valence-arousal labels.
• Developing personalized music recommendation systems based on predicted emotional response.
• Analyzing the relationship between acoustic features and perceived emotion for music production.
• Supporting psychotherapeutic or market analysis applications through affective computing.

Strengths

• Annotations by 30 music experts, which likely enhances label reliability.
• Contains 2,496 instrumental songs, providing a substantial corpus for model training.
• Dataset size of 17.1 GB indicates a significant volume of raw audio data.
• Structured into raw audio, annotations, and features folders for organized access.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for specific modeling tasks.
• Data may reflect bias inherent to the source platform and the specific instrumental music collection.

Provenance

Source

figshare, author Tong Zhang

Collection Method

Instrumental pieces annotated by 30 music experts using a methodology grounded in previously validated studies.

Time Range

null

Freshness

Last updated 2026-04-20 07:58:59

Geography

null