Analysis data and code for Screening bird recordings with species recognition scores: vocalization-type classification and field acoustic-type coverage

Analysis data and code accompanying the manuscript ‘Screening bird recordings with species recognition scores: vocalization-type classification and field acoustic-type coverage’ by YuanZhe Cui. The study evaluates how fixed BirdNET species-score screening rules affect the preservation and downstream use of annotated bird vocalizations. The archive contains three components: (1) zebra finch reference-library analyses and a Tawny Pipit song comparison; (2) an external adult zebra finch context and focal-removal diagnostic; and (3) a Dupont’s lark field reanalysis covering 288 recordings, including a whole-S125-recording exclusion sensitivity. Included materials comprise analysis code, selected scalar scores and acoustic features, source annotations and metadata, fixed protocols, reference result tables, provenance hashes, attribution, and numerical reanalysis instructions. Original or converted audio, model weights and the manuscript are not included. These reduced inputs support the documented fixed-rule numerical analyses, not repetition of neural inference or independent validation of biological annotations. Source datasets and model versions are identified in the component documentation. Bird- or population-balanced summaries preserve the analysis units defined for each component; the data do not support estimates of natural noise-rejection precision or actual human labour savings. This package was prepared from completed analyses. Packaging and privacy-path changes received static and file-integrity checks; the sanitized candidate was not numerically replayed. Included successful replay records describe earlier archive versions and retain that historical scope. AI assistance with implementation, analysis, checks and writing is disclosed in the accompanying materials. Third-party materials retain their own attribution and terms.