Annotated Amazon Soundscapes: Bird Vocalizations from Peru

Twenty-one hour-long soundscape recordings from the Southwestern Amazon Basin contain 14,798 bounding box annotations for 132 bird species. The data was collected in 2019 at the Inkaterra Reserva Amazonica in Peru using ten SWIFT recording units. It was partially featured as test data in the 2020 BirdCLEF competition for machine learning algorithm training and evaluation.

Use Cases

• Training bird species detection models based on annotated bounding boxes in audio spectrograms.
• Assessing spatio-temporal variation in avian vocal activity across different forest habitats (intact, degraded, edge).
• Benchmarking audio event detection algorithms using a dataset from a recognized competition (BirdCLEF 2020).
• Studying the impact of environmental factors like deforestation on biodiversity using acoustic indices.

Strengths

• Contains a high volume of detailed annotations with 14,798 bounding box labels.
• Covers a diverse set of 132 bird species from a biodiverse region.
• Provides detailed technical metadata on recording equipment, including microphone sensitivity (-44 dB) and sampling rate (48 kHz).
• Data collection methodology is explicitly described, including deployment across ten sites in distinct forest habitats.

Limitations

• Key metadata such as column names, row counts, and exact file size are not provided in the available sources.
• The dataset's last update date is unknown, which may affect perceived freshness.
• The microphone frequency response is assumed, not measured, for the range 100 Hz to 7.5 kHz.

Provenance

Source

W. Alexander Hopping, Cornell University / K. Lisa Yang Center for Conservation Bioacoustics.

Collection Method

Acoustic data collected continuously with ten omnidirectional SWIFT recorders set to record uncompressed WAVE files.

Time Range

January 14 to February 2, 2019 (rainy season).

Geography

Inkaterra Reserva Amazonica, a 2 km² lowland rainforest reserve on the Madre de Dios river, Madre de Dios, Peru.