Real Binaural Speech Source Directivity Dataset

This dataset contains 184 real binaural speech recordings generated for the study of speech source directivity classification. The dataset is designed to support the binary classification of whether a speech source is oriented toward or away from a listener. The language is chilean Spanish. The recordings were generated under controlled acoustic and spatial conditions using Roland CS-10EM binaural microphone/earphones and ZOOM H4n Pro recorder. Each audio file corresponds to a binaural recording of a speech source with associated spatial, orientation and distance metadata. Each audio file is a binaural speech signal with two channels: left channel right channel The dataset contains real speech stimuli under different combinations of source position, direction of arrival, source orientation, and distance. The speech source directivity classification task is formulated as a binary classification problem: Label Meaning 0 Reversed / not facing the listener 1 Facing the listener The label is derived from the relative orientation of the speech source with respect to the listener. This dataset is intended for research on: speech source directivity classification binaural audio classification spatial audio analysis acoustic simulation machine listening direction-aware audio systems assistive listening and hearing-aid-related applications This dataset is focused on source directivity classification, not on automatic speech recognition, speaker identification, or general-purpose sound event classification. Although spatial metadata are included, the primary target is the binary classification of speech source orientation relative to the listener.