Birdcall tagging dataset

This is a dataset of birdcall recordings, a specific subset gathered from the xeno-canto collection to form a manually audio-tagged dataset. This dataset aims to identify birdcall against background noise / other sound types. Though the dataset can be used to tackle birdcall species classification problem, all species data should be considered weakly labelled.

Audios were tagged using the following rules:

1. A birdcall will occur within 500ms of start time and 500ms before the end time.
2. Any positive tag will not contain more than 1s of background noise between birdcalls.
3. Each record represents a single bird species unless multiple birdcalls (multiple bird species or different species) overlapping

Dataset currently consists of 207 records from 13.5 minutes of birdcall recordings with four bird species. (Update every two to three days)

Dataset

The following is a detailed description of the fields in the dataset:

• id : Index number of records by country
• common_name : English name of the species
• scientific_name : Scientific name of the species
• subspecies : Subspecies name of the species
• recordist : Name of recordist
• date : Date of the recording was made
• time : Time of the recording was made
• location : Location of the recording was made
• country : Country of the recording was made
• latitude : Latitude of the recording in decimal coordinates
• longitude : Longitude of the recording in decimal coordinates
• elevation : Elevation of the recording was made (in meters)
• song_type : Sound type of the recording (e.g. 'call', 'song', etc)
• remarks : Additional remarks by recordist (Line separator ';')
• back_latin : List of scientific name of background species
• catalogue_number : Catalogue number of the recording on xeno-canto
• rating : Rating of the recording (The best A to the worst E)
• file_type : File extension
• file_length : Lenth of sound file in seconds
• sampling_rate : Sample rate of the recording file
• bitrate : Bit rate of the recording file
• channels : Number of channels used to record the file
• bird-seen : Was the recorded bird visually identified? (yes/no)
• playback-used : Was playback used to lure the bird? (yes/no)
• volume : Sound characteristics of birdcall
• speed : Sound characteristics of birdcall
• pitch : Sound characteristics of birdcall
• length : Sound characteristics of birdcall
• num_notes : Sound characteristics of birdcall
• variable : Sound characteristics of birdcall
• license : License of the recordings
• start_time : Start time of tag in seconds (round down to nearest 2 decimal places)
• end_time : End time of the tag in seconds (round up to nearest 2 decimal places)
• primary_target : Primary target of recording (True/False)

Project Structure

.
├── ...
├── United_States           # Name of country
│   ├── United_States_1.csv # Dataset 1
│   ├── United_States_2.csv # Dataset 2 
│   └── ...                 # etc.
└── ...

License

This project follows CC-BY-NC-4.0. Thus, the dataset is freely available for academic or individual research, but is restricted for commercial use.

Recordings © the recordist. See recording details within the dataset for license information.

Acknowledgements

The data recordings in the dataset were collected by various birding enthusiasts and uploaded to and stored by xeno-canto: xeno-canto. If you make use of these recordings in your work, please cite the specific recording and include acknowledgement of and a link to the xeno-canto website.

Warning

Xeno-canto API does not allow heavy crawling. Please be mindful when accessing xeno-canto resources.

For more information, please refer to this page.