README.md

metadata

license: apache-2.0
task_categories:
  - audio-classification
  - audio-to-audio
  - text-to-audio
task_ids:
  - multi-class-classification
tags:
  - audio
  - sound
  - classification
  - music
  - source-separation
  - audio-generation
  - sound-synthesis
size_categories:
  - 100k<n<1M
dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: label
      dtype:
        class_label:
          names:
            - kick
            - snare
            - hihat
            - clap
            - 808
            - perc
            - fx
            - cymbal
            - vox
            - texture
            - melodic
    - name: duration
      dtype: float32

Dataset Name

drums is a dataset of more than 150,000 samples, including drums and various percussion, melodic and vocal samples

Dataset Description

This dataset contains 154,600 short audio files with duration up to 2 seconds each, labeled with sound categories. The dataset is designed for classification, separation, and generation tasks.

Key Features

• Dataset size: 154,600 audio files
• Audio duration: [0.05,2] seconds
• Audio format: wav
• Sample rate: 44100 Hz
• Classification type: Multiclass

Class Labels

The dataset contains the following sound categories:

• "perc"
• "kick"
• "vox"
• "melodic"
• "clap"
• "snare"
• "drumfill"
• "texture"
• "808"
• "bass"
• "cymbal"
• "hihat"
• "fx"