How karolpiczak/esc-50 works
Key Components to Include:
Data Organization Layer
• Audio Files Repository
• Hierarchical structure of WAV files (5-second, 44.1 kHz, mono)
• Naming convention pattern: {FOLD}-{CLIP_ID}-{TAKE}-{TARGET}.wav
• Metadata Layer
• CSV file structure (esc50.csv)
• Human evaluation data (esc50-human.xlsx)
Dataset Structure
• 5 Cross-validation Folds
• 50 Semantic Classes
• 2000 Audio Recordings
• ESC-10 Subset (10 selected classes)
Classification Categories
• Animals
• Natural soundscapes & water sounds
• Human, non-speech sounds
• Interior/domestic sounds
• Exterior/urban noises
Use a Layered Architecture
┌─────────────────────────────────────────┐
• ESC-50 Dataset │
• ┤
• ┌─────────────┐ ┌────────────┐ │
• Audio Data │ │ Metadata │ │
• ┘ └────────────┘ │
• ┌─────────────┐ ┌────────────┐ │
• 5 Folds │ │ CSV │ │
• ┘ └────────────┘ │
• ┤
• Classification Categories │
• ┘
• Cardinality (1:n, n:1, etc.)
Additional Elements
• Data format specifications
• Cross-validation fold structure
• File naming conventions
Technical Specifications
• Audio format: WAV (5 seconds, 44.1 kHz, mono)
• Dataset size: 2000 recordings
• Cross-validation: 5 folds
• Class distribution: 40 examples per class
License Information
• Creative Commons Attribution Non-Commercial
• CC BY for ESC-10 subset
The distinction between ESC-50 and ESC-10 subsets
This system design will help developers understand:
• How to navigate the dataset
• The relationship between different components
• The data organization principles
• The classification structure
• How to properly use the cross-validation folds
Remember to maintain a balance between detail and clarity, as this is primarily a dataset project rather than a software system.