nkdem/HEAR-DS-16k

HEAR-DS Background Audio (16kHz)

Binaural background audio recordings from the HEAR-DS (Hearing Aid Research Database of Sounds) dataset, downsampled to 16kHz and chunked into 10-second segments for speech enhancement and acoustic scene classification research.

Dataset Description

This dataset contains background noise recordings from 7 acoustic environments, captured using in-the-canal (ITC) hearing aid microphones. Each sample includes stereo (left/right ear) audio.

Environments

Split	Environment	Samples
cocktail_party	Multi-speaker social settings	716
in_traffic	Road traffic noise	1,000
in_vehicle	Inside vehicles (car, bus, train)	1,094
music	Various music genres	2,991
quiet_indoors	Low-noise indoor environments	951
reverberant	Reverberant spaces	1,007
wind_turbulence	Outdoor wind noise	1,034
Total		8,793

Features

• id: Unique sample identifier
• audio_left: Left ear ITC microphone recording (16kHz, mono)
• audio_right: Right ear ITC microphone recording (16kHz, mono)
• environment: Acoustic environment category
• rec_id: Original recording session ID
• cut_id: Cut/segment ID within recording
• snip_id: 10-second chunk index

Usage

from datasets import load_dataset

# Load all environments
ds = load_dataset("nkdem/HEAR-DS-16k")

# Load specific environment
traffic = load_dataset("nkdem/HEAR-DS-16k", split="in_traffic")

# Access a sample
sample = traffic[0]
print(sample["environment"])  # "InTraffic"
print(sample["audio_left"]["array"].shape)  # (160000,) - 10 seconds at 16kHz

Processing Details

• Original sample rate: 48kHz
• Target sample rate: 16kHz (downsampled with librosa)
• Chunk duration: 10 seconds (160,000 samples)
• Channels: Mono per ear (stereo pair preserved as separate columns)
• Microphone: In-The-Canal (ITC) hearing aid microphones

Citation & Attribution

This dataset is derived from HEAR-DS, created by Hörzentrum Oldenburg:

Hohmann, V., et al. "The HEAR-DS database of acoustic scenes and events for hearing aid research." Hörzentrum Oldenburg gGmbH.

Original source: https://www.hz-ol.shop/en/hear-ds.html

If you use this dataset, please cite the original HEAR-DS database and acknowledge Hörzentrum Oldenburg.

Intended Use

This dataset is intended for:

• Speech enhancement model training (as noise source for augmentation)
• Acoustic scene classification research
• Hearing aid algorithm development
• Audio machine learning research

Licence

This is a processed version of HEAR-DS for research purposes. Please refer to the original HEAR-DS page for licensing terms. Some components may have additional restrictions from third-party sources (CHiME, GTZan).