MAESTRO Real: 49 Audio Files with Crowdsourced Strong Labels

49 real-life audio recordings totaling 189 minutes and 52 seconds, sourced from the TUT Acoustic Scenes 2016 dataset. The data includes soft labels representing estimated strong labels derived from crowdsourced annotations via Amazon Mechanical Turk. The dataset was created by Irene Martín-Morató of Tampere University for studying label estimation from crowdsourcing.

Use Cases

• Training models for acoustic scene classification based on five real-world scenes: cafe/restaurant, city center, grocery store, metro station, and residential area.
• Developing algorithms to estimate strong labels from uncertain, crowdsourced annotations using the provided soft labels.
• Benchmarking signal processing techniques for audio event detection within complex, real-life acoustic environments.

Strengths

• Contains 49 real-life audio recordings with a total duration of 189 minutes and 52 seconds.
• Provides soft labels with values between 0 and 1, quantifying annotator uncertainty for 17 event classes across 5 acoustic scenes.
• Audio files are a curated subset from the established TUT Acoustic Scenes 2016 research dataset.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• Row count and file formats are unknown, which may limit suitability assessment.

Provenance

Source

Tampere University, derived from the TUT Acoustic Scenes 2016 dataset.

Collection Method

Audio files are a subset of real-life recordings; annotations were crowdsourced via Amazon Mechanical Turk.