MyGardenBird: A Machine-Learning-Ready Bird Sound Dataset for Twelve Common Malaysian Birds

MyGardenBird is a machine-learning-ready bioacoustic dataset of twelve common bird species from Peninsular Malaysia and the wider Indo-Malayan region. The primary release contains 7,200 manually reviewed three-second audio clips (600 per species, 6.0 h total) sourced from Xeno-canto, sampled at 16 kHz and stored as 16-bit PCM mono WAV. A parallel 44.1 kHz subset (6,950 clips) is included for applications requiring higher spectral resolution. Each clip carries per-sample metadata: Xeno-canto provenance, geographic coordinates, vocalisation type, and estimated SNR. The 1,381 source recordings are partitioned into train/validation/test sets (80:10:10) at the source level using mixed-integer programming to prevent data leakage. BirdNET zero-shot validation yields 97.94% accuracy at 16 kHz (98.06% at 44.1 kHz); CNN baselines across three architectures achieve 92–96% test accuracy at 16 kHz. Related resources Paper: MyGardenBird: A Machine-Learning-Ready Bird Sound Dataset for Twelve Common Malaysian Birds — Zabidi, Idris MY, Idris N — https://arxiv.org/abs/2605.20853 Curation code: https://github.com/mun3im/mygardenbird.