The dataset is currently empty. Upload or create new data files. Then, you will be able to explore them in the Dataset Viewer.

Infant Cry Detection Dataset — 50+ Hours of Real Baby Cry Audio

50+ hours of real infant cry recordings for training infant cry detection, cry classification, and sound event detection models. Manually verified files captured in natural domestic conditions, with per-file metadata on location, background noise, and recording device. 7× larger than the most cited academic infant cry corpus (CryCeleb, 6.5h) — and licensed for commercial use.

Contact us and share your feedback — receive additional samples for free! 😊

Key Highlights

  • 50+ hours of real-world infant cry audio
  • Manually verified — every recording reviewed for clear cry audibility
  • Authenticity filter applied — files resembling internet downloads removed
  • Indoor + outdoor capture conditions
  • Background noise variation — quiet, moderate, and noisy environments
  • No synthetic audio, no augmentation, no AI-generated content
  • Smartphone-first recordings (matches baby monitor deployment conditions)
  • Commercial license available — full version cleared for production use

Use This Dataset For

  • Infant cry detection — train binary or multi-class cry detectors for smart baby monitors, IoT cameras, and nursery devices
  • Sound event detection (SED) — cry as a target class in AudioSet-style models
  • Sleep tracking applications — detect cry events to estimate infant sleep quality
  • Cry-aware smart home automation — trigger lighting, audio response, or parent notifications
  • Parental support apps — distinguish cry from other infant sounds (cooing, babbling, fussing)
  • Audio classification research — clean, labeled training data on a focused class

Dataset Structure

infant-cry-detection-dataset/
├── audio/
│   ├── cry_00001.wav
│   ├── cry_00002.m4a
│   └── ... (WAV + M4A files)
├── metadata.csv
└── README.md

metadata.csv schema

Field Type Values
record_id string unique recording ID
file_name string path to audio file
file_ext string .wav, .m4a
duration_sec float 10–100 seconds
sample_rate_hz int 48000 (majority), 44100, 16000
validation_status string accepted
signal_clear string yes (all files confirmed clear)
recording_location string indoor, outdoor
background_noise_level string quiet, moderate, noisy
device_type string smartphone, laptop, tablet, external_mic
codec_name string pcm_s16le, aac

Dataset Statistics

Metric Value
Total duration 50+ hours
File formats WAV (majority) + M4A
File duration range 10–100 sec
Sample rates 48 kHz / 44.1 kHz / 16 kHz
Recording locations indoor + outdoor
Noise conditions quiet / moderate / noisy
Capture devices smartphone (majority), laptop, tablet, external mic
Verification manual review of every file

How This Compares to Academic Infant Cry Datasets

Dataset Duration License Best for
Axon Labs Infant Cry Detection 50+ hours Commercial Production cry detection & SED
CryCeleb (Ubenwa) 6.5 hours CC-BY-NC-ND (research only) Speaker verification across infant identities
Donate a Cry Corpus <1 hour Open, crowdsourced Cry cause classification (small-scale research)

CryCeleb and Donate a Cry remain the leading academic resources for their respective research tasks. Our dataset complements them by providing the scale and licensing required for production-grade detection models in commercial baby monitor, sleep tracking, and smart home applications.

Quick Start — Loading with 🤗 Datasets

from datasets import load_dataset

dataset = load_dataset("AxonData/infant-cry-detection-dataset")
print(dataset)

sample = dataset["train"][0]
print(sample["audio"])                    # audio array + sampling_rate
print(sample["recording_location"])       # e.g. "indoor"
print(sample["background_noise_level"])   # e.g. "quiet"
print(sample["device_type"])              # e.g. "smartphone"

Quick Start — PyTorch DataLoader

import torch
from datasets import load_dataset

ds = load_dataset("AxonData/infant-cry-detection-dataset", split="train")

def collate(batch):
    waveforms = [torch.tensor(item["audio"]["array"]) for item in batch]
    metadata = [{
        "location": item["recording_location"],
        "noise": item["background_noise_level"]
    } for item in batch]
    return waveforms, metadata

loader = torch.utils.data.DataLoader(ds, batch_size=8, collate_fn=collate)

Sample vs Full Version

This HuggingFace repository contains a sample subset for evaluation. The full 50+ hour dataset is licensed for commercial use through Axon Labs.

Full version of dataset is available for commercial usage — leave a request on our website Axonlabs to purchase the dataset 💰

What Makes This Dataset Unique

  • Largest infant cry audio corpus available commercially — multiple times the size of leading academic alternatives
  • Manually verified, not scraped — every file reviewed; suspicious files removed
  • Authenticity-filtered — directly addresses a documented problem in open datasets (noisy, internet-sourced contamination)
  • Real smartphone recordings — matches deployment conditions for baby monitor apps and smart home devices
  • Backed by a biometric AI specialist — Axon Labs builds datasets used by 21% of iBeta 2025 certified companies
  • GDPR-compliant, ethically sourced — explicit parental consent for all recordings

Two Dataset Versions Available

  • Sample Version — open subset for EDA, evaluation, and proof-of-concept (this repo)
  • Full Version — 50+ hours of audio with complete metadata, licensed for commercial training

Contact us to choose the version that fits your project.

FAQ

Q: How large is this dataset compared to public alternatives? The full version is 50+ hours of curated infant cry audio. The most widely cited academic alternative, CryCeleb, contains 6.5 hours of cry expirations. Donate a Cry Corpus contains less than 1 hour of crowdsourced audio with documented quality issues. By volume of verified cry audio licensed for commercial training, this is the most extensive resource currently available.

Q: Can I use this dataset to train a production baby monitor or sleep tracking app? Yes — the full commercial version is licensed exactly for this. The sample version on HuggingFace is for evaluation, EDA, and proof-of-concept work. For production training, request access to the full version.

Q: What metadata is provided? Recording location (indoor/outdoor), background noise level (quiet/moderate/noisy), recording device type (smartphone/laptop/tablet/external microphone), validation status, and per-file signal clarity confirmation. All files in the curated set have been confirmed as clear, original infant cry recordings.

Q: Are there labels for cry cause (hunger, pain, discomfort)? Not in this release. The dataset is designed for cry detection (is a cry present) and sound event detection use cases. If you need labeled cry-cause data, we can collect a custom dataset on request through our data collection service.

Q: Is the data ethically collected? Yes. All recordings were captured with explicit parental consent and documented chain of custody. Processing follows GDPR and applicable standards for audio data involving minors. Full ethical sourcing documentation is available for the commercial version.

Citation

If you use this dataset in your research, please cite:

@misc{axonlabs2026infantcry,
  title  = {Infant Cry Detection Audio Dataset},
  author = {Axon Labs},
  year   = {2026},
  url    = {https://axonlab.ai/dataset/infant-cry-dataset/}
}

keywords: infant cry dataset, baby cry dataset, infant cry detection, infant cry classification, baby cry detection, sound event detection, audio classification dataset, baby monitor dataset, smart home audio, real-world infant audio, parental tech dataset

Visit us at Axonlabs to request a full version of the dataset for commercial usage.

Downloads last month
-