Non-Semantic Human Acoustic Transients for Large-Scale Data Analytics (CC0)

This dataset contains over 7,500 short-duration human-generated acoustic transient recordings collected over a period of approximately 37 months. The recordings capture non-semantic acoustic events and are intended to support experimentation and research in audio signal processing, machine learning, and data analytics.All recordings are provided in uncompressed WAV format. Audio was captured using a mobile voice recording application across a range of everyday environments, resulting in natural variation in background noise, volume, and acoustic conditions. Each recording was manually reviewed, and samples with excessively low signal level, dominant background noise, or non-target artefacts were removed. The dataset therefore retains realistic variability while excluding unusable recordings.No preprocessing has been applied to the audio files in order to preserve flexibility for downstream tasks. Users may wish to apply standard audio preprocessing techniques, such as trimming to a fixed duration, normalisation, or time–frequency transformation, depending on their application.Files are named sequentially using integer identifiers. File order does not encode temporal or categorical information.Citation of this Figshare record is appreciated when used in academic work.