Annotated Bird Soundscapes from Colombian and Costa Rican Coffee Farms

34 hour-long soundscape recordings from neotropical coffee farms in Jardín, Colombia and San Ramon, Costa Rica, collected in 2019. Expert ornithologists provided 6,952 bounding box labels for 89 different bird species. The collection was created by Álvaro Vega-Hidalgo of Cornell University and partially featured in the 2021 BirdCLEF competition.

Use Cases

• Training bird call detection models based on bounding box annotations in time and frequency.
• Evaluating machine learning algorithms for species identification using expert-labeled audio.
• Monitoring the impact of farm management practices (e.g., pesticide application, pruning) on bird activity.
• Correlating different types of tree cover on coffee farms with the presence of target bird species.

Strengths

• 6,952 expert-annotated bounding box labels provide a substantial training signal.
• Covers 89 different bird species, indicating taxonomic diversity.
• Recordings are 34 hours long and capture two daily avian activity peaks (dawn and dusk).
• Data collection protocol is detailed, specifying recorder type (SWIFT), height (3m), and audio processing (resampled to 32 kHz FLAC).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic bias inherent to the two specific farm locations in Colombia and Costa Rica.

Provenance

Source

Cornell University

Collection Method

Recordings made using SWIFT recorders positioned 3m above ground, capturing one-hour sound files during dawn and dusk avian activity peaks. Annotations were created in Raven Pro by expert ornithologists.

Time Range

2019

Geography

Coffee farms in Jardín, Colombia and San Ramon, Costa Rica