Dataset Viewer
Auto-converted to Parquet
Search is not available for this dataset
audio
audioduration (s)
1.47
47.4
label
class label
0 classes
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
null
End of preview. Expand in Data Studio

🐾 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.

πŸ“‚ Folder Structure

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

Downloads last month
342