Back to datasets
Dataset assetOpen Source CommunityAudio Classification

renumics/esc50

ESC‑50 is an environmental sound classification dataset comprising 50 distinct sound categories such as animal noises (dog, cat, chicken), natural sounds (rain, sea waves, wind), human sounds (laughter, cough, footstep), and mechanical sounds (helicopter, chainsaw, siren). Features include audio files, labels, and fold information. The training set contains 2,000 samples (≈882 MB). The dataset is released under a Creative Commons Attribution‑NonCommercial license.

Source
hugging_face
Created
Nov 28, 2025
Updated
Nov 9, 2023
Signals
327 views
Availability
Linked source ready
Overview

Dataset description and usage context

Dataset Overview

Dataset Information

Features

  • src_file: string
  • fold: int64
  • label: categorical with 50 classes (e.g., 0: dog, 1: rooster, …, 49: hand_saw)
  • esc10: boolean
  • take: string
  • audio: audio data

Data Split

  • train: 2,000 samples, 882,179,256 bytes

Size

  • Download size: 773,038,488 bytes
  • Dataset size: 882,179,256 bytes

Configuration

  • default: data files located at data/train-*

License

  • Creative Commons Attribution‑NonCommercial (cc‑by‑nc‑2.0)

Task Type

  • Audio Classification

Scale

  • 1K < N < 10K
Need downstream help?

Pair the dataset with AI analysis and content workflows.

Once the source passes your review, move straight into summarization, transformation, report drafting, or presentation generation with the JuheAI toolchain.

Explore AI studio