🐾 Dog Vocal Detection

This dataset is curated from internet videos to support research in dog vocalization detection using both weak and strong supervision.

• It contains approximately 7,500 seconds of strongly labeled training audio
• Over 9,000 seconds of weakly labeled clips sourced from AudioSet are included.
• The dataset also provides 24 hours of unlabeled audio clips from our own collection.
• To simulate realistic conditions, some clips feature dogs present without barking, reflecting challenging real-world scenarios.

A corresponding validation and test set are available as part of the Barkopedia Dog Vocal Detection Challenge on Hugging Face.

📂 Folder Structure

.
├── strong_audios
│   ├── UC_Aswv9nASkcjXoXKPVkPiQ+_mMhlYEgxoE+00002_pitbull.wav
│   ├── UC_Aswv9nASkcjXoXKPVkPiQ+3sZLaAgADuo+00002_pitbull.wav
│   ├── UCeqTOU_4b6UEmPgBdqd4YWg+rwo5lVV3N2I+00000_chihuahua.wav
│   ├── ...
│   └── UCze-YohIl415Ric6I2SN0ow+SoYhkuSXOZo+00007_shiba_inu.wav
│       
├── strong_audios_tsv
│   └── train.tsv
│
├── unlabeled_audios
│   ├── chihuahua
│       ├── UC_WHiJEE5B-QOKpwLyaNcXg+E0VNPqB7c7g+00001.wav
│       ├── ...
│       └── UC0Qw7Ew6lSPHfAm83sRu_3g+QcHYPW9irkE+00000.wav
│   ├── german_shepherd
│   ├── husky
│   ├── labrador
│   ├── pitbull
│   └── shiba_inu
│
└── weak_audios
    ├── chihuahua
    ├── german_shepherd
    ├── husky
    ├── labrador
    ├── pitbull
    └── shiba_inu

where the train.tsv takes a format like so:

UCkMjjl3JEZpp4sRUJx5eJEA+X-X8s0DXXI8+00004_chihuahua.wav	9.59548	10.01272	dog
UCkMjjl3JEZpp4sRUJx5eJEA+X-X8s0DXXI8+00004_chihuahua.wav	10.30567	11.57513	dog
UCvHJzqB1KZMpZniAVWy3tkA+fSQeU1uAQbc+00001_chihuahua.wav	0.76266	6.85668	dog_noise
UCvHJzqB1KZMpZniAVWy3tkA+fSQeU1uAQbc+00001_chihuahua.wav	7.5288	9.00958	dog
...

Specifically, each row should map one wav file from the test set to the onset and offset (in seconds) of a dog or dog_noise event. In particular, a dog event is a continuum of dog bark(s) that lacks any noticable time gaps. A dog noise event is a dog event that occurs simultaneously with noticable noise (eg, human speech or environmental sounds).

🏎️ Challenge

This training set has been released for use in the Barkopedia Dog Vocal Detection Challenge on Hugging Face.