MAESTRO Synthetic: 20 Synthetic Audio Files with Crowdsourced Labels

20 synthetic audio files, each 3 minutes long, created for studying strong label estimation from crowdsourcing. The dataset includes reference annotations, estimated strong labels from Amazon Mechanical Turk, and weak labels for 10-second segments. It was created by Irene Martín-Morató of Tampere University for a 2021 WASPAA paper.

Use Cases

• Developing label aggregation algorithms based on crowdsourced weak labels mentioned in the description.
• Benchmarking sound event detection models using the provided synthetic audio and reference annotations.
• Studying the relationship between weak and strong labels for audio segments as described.
• Training models for audio tagging using the provided segment-level weak labels.

Strengths

• Includes 20 distinct audio files, each with a defined 3-minute duration.
• Provides multiple annotation layers: ground truth, estimated strong labels, and segment-level weak labels.
• Synthetic nature allows for controlled study of label estimation without real-world recording noise.

Limitations

• Dataset scale is small with only 20 audio files; row count is unknown.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

Tampere University

Collection Method

Audio files were synthetically generated using Scaper, with annotations crowdsourced via Amazon Mechanical Turk. Source audio excerpts are from the Urban Sound 8k dataset via freesound.org.