Bird Audio Classification Splits & Mel\u2011Spectrograms (Xeno\u2011canto A\u2013M, v1.0)

We release 2\u202f984 twenty\u2011second mono bird\u2011song clips drawn from the public Xeno\u2011Canto Bird Recordings Extended (A\u2013M) corpus. After mirroring the original MP3s and metadata, we curate the 30 most\u2011represented species (rating\u202f\u2265\u202f3.0, duration\u202f\u2264\u202f300\u202fs) and convert each clip to WAV plus a noise\u2011reduced counterpart. Three machine\u2011learning views are provided: (1) 128\u202f\u00d7\u202f432 Mel\u2011spectrograms (npy), (2) YAMNet log\u2011Mel tensors and 1\u202f024\u2011D embeddings (npy), and (3) 38 handcrafted acoustic features (csv). Stratified 60\u202f\/\u202f20\u202f\/\u202f20 train\u2011validation\u2011test splits and label files are included. The resource supports fine\u2011grained bio\u2011acoustic classification, transfer\u2011learning benchmarks and classical\u2011vs\u2011deep pipelines. All preprocessing scripts are open\u2011sourced for full reproducibility under MIT licence; original recordings retain their Xeno\u2011Canto attribution requirements.