Synthetic Data by Construction

This section demonstrates the construction of synthetic audio data with controlled properties. Two classes are created by combining four distinct audio concepts in different ways, allowing for ground-truth evaluation of DRSA's concept extraction capabilities.

Class 1

Class 1 is constructed by combining four basic audio concepts. Listen to each individual concept and the final combined sample.

Class 1 construction visualization
Fig. 1: Visualization of Class 1 construction: Four audio concepts are combined to create the final synthetic sample.

Audio Components

Listen to each individual concept and the final combined sample:

Concept 10:00 / 0:00
Concept 20:00 / 0:00
Concept 30:00 / 0:00
Concept 40:00 / 0:00
Final Sample0:00 / 0:00

Class 2

Class 2 uses the same four concepts but with different combinations and parameters, creating a distinct class for classification.

Class 2 construction visualization
Fig. 1: Visualization of Class 2 construction: The same concepts combined differently to create a separate class.

Audio Components

Listen to each individual concept and the final combined sample:

Concept 10:00 / 0:00
Concept 20:00 / 0:00
Concept 30:00 / 0:00
Concept 40:00 / 0:00
Final Sample0:00 / 0:00