DCASE Challenge Dataset
The DCASE Challenge Dataset brings together audio recordings from a variety of everyday environments. Designed specifically for training and evaluating sound scene identification models, this dataset is used as a reference in international competitions dedicated to acoustic context analysis.
Several thousand audio recordings, WAV format
Free for academic and research use (DCASE specific license)
Description
The dataset includes sounds captured in real conditions, including:
- Public spaces (parks, streets, stations,...)
- Indoor spaces (cafes, offices, classrooms,...)
- Typical domestic or urban scenes
- From different countries and cultural contexts
Each recording is accurately annotated for immediate use in supervised audio classification tasks.
What is this dataset for?
The DCASE Challenge Dataset is used primarily for:
- Training audio scene recognition models (Soundscape Classification)
- The development of robust algorithms for identifying the acoustic context
- Comparative evaluation (benchmark) of performances between acoustic or multimodal approaches
- The creation of intelligent acoustic monitoring systems (connected cities, public spaces, IoT...)
Can it be enriched or improved?
Yes, for example:
- By integrating other similar audio corpora (UrbanSound8K, ESC-50, AudioSet)
- By creating more complex acoustic contexts via superposition or simulation of mixed environments
- By enriching the annotations with additional contextual metadata (time, weather, type of audience...)
- By testing recognition scenarios under difficult conditions (low audio quality, loud noise, etc.)
🔗 Source: DCASE Challenge Dataset
Frequently Asked Questions
Is the dataset suitable for commercial use?
Not directly. Its use is limited to research and participation in DCASE challenges, unless otherwise stated on the official website.
Are there different editions of the DCASE dataset?
Yes, the dataset evolves every year with new editions offering varied sound environments and adapted test scenarios.
How do I access detailed annotations?
Precise annotations are provided with the dataset download from the official DCASE website, making it easy to integrate it into training pipelines.
