A behaviourally normed database of 1,377 natural sounds for auditory cognition and neuroscience

The MaMa Sounds database, a compilation of 1,377 natural sounds, has been developed to facilitate auditory cognition and neuroscience research. This extensive collection, comprising two-second sounds representing 240 distinct source-action classes, marks a collaborative effort between academic teams in Maastricht and Marseille.

Each sound in the database has been meticulously curated and segmented, sampled at a rate of 16 kHz, and accurately labeled with a noun identifying the sound source and a verb describing the action. The dataset is enriched with comprehensive deidentified trial-level data, including identification accuracy, confidence, agreement, and familiarity metrics.

Furthermore, the database incorporates multiple per-sound norms, derived through principal component analysis, to offer valuable insights into sound identification and familiarity.

The database provides a rich set of norms, including noun, verb, and joint noun-verb norms, presented as direct means and medians along with the number of contributing observations. These norms preserve process-specific information, while two principal-component scores offer compact overall behavioral-identifiability measures. These measures are derived from response ease, semantic correspondence, agreement, and familiarity, providing a concise summary of the data.

In addition to the sound data and norms, the MaMa Sounds repository also includes deterministic response-cleaning code, participant and reference Word2Vec representations, and code to reproduce public sound-level tables. This comprehensive resource is designed to support various applications in auditory cognition and neuroscience, such as stimulus selection, matching, and continuous modeling.