Datasets:

sguajardo799
/

simulated-binaural-speech-directivity

This dataset contains 96,000 simulated 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 recordings were generated under controlled acoustic and spatial conditions using an acoustic simulation workflow based on RAVEN (https://www.virtualacoustics.org/). Each audio file corresponds to a binaural rendering of a speech source with associated spatial, orientation, distance, reverberation, and intensity metadata.

Authors

Rhoddy Viveros-Muñoz¹ and Sebastian Guajardo-Herrera¹

¹ Departamento de Electrónica e Informática, Universidad Técnica Federico Santa María, Concepción, Chile.

Dataset Overview

The dataset addresses the problem of speech source directivity classification. Unlike conventional sound source localization tasks, which focus mainly on estimating where a sound comes from, this dataset focuses on whether the speech source is facing or not facing the listener.

This distinction is relevant for spatial audio analysis, auditory scene understanding, assistive listening technologies, hearing-aid-oriented machine listening, and intelligent acoustic systems.

Dataset Structure

The dataset follows an audiofolder-compatible structure:

dataset/
├── data/
│   ├── OrientationData_HRTF00001.wav
│   ├── OrientationData_HRTF00002.wav
│   ├── OrientationData_HRTF00003.wav
│   └── ...
└── metadata.csv

The data/ directory contains the binaural .wav files.
The metadata.csv file contains one row per audio file.

Audio Data

Each audio file is a binaural speech signal with two channels:

left channel
right channel

The dataset contains simulated speech stimuli rendered under different combinations of source position, direction of arrival, source orientation, distance, reverberation time, and intensity.

Metadata

The metadata.csv file contains the following columns:

Column	Description
`file_name`	Relative path to the audio file. Example: `data/OrientationData_HRTF00001.wav`
`x (m)`	Source position on the x-axis, in meters
`y (m)`	Source position on the y-axis, in meters
`z (m)`	Source position on the z-axis, in meters
`DoA (deg)`	Direction of arrival of the source, in degrees
`angle source orientation (deg)`	Absolute source orientation angle, in degrees
`distance (m)`	Source-listener distance, in meters
`Rev Time`	Reverberation time condition
`Intensity (%)`	Relative source intensity
`angle_source`	Relative source orientation angle used for directivity labeling
`classification`	Binary directivity label

Labels

The target variable is provided in the classification column.

The 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.

Acoustic and Spatial Conditions

The dataset includes systematic variation in:

source position
direction of arrival
source orientation
source-listener distance
reverberation time
relative source intensity

The direction of arrival covers the horizontal plane using multiple azimuthal positions. Source orientations cover the full angular range, allowing the dataset to represent both facing and reversed speech directivity conditions.

Intended Use

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

Scope

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.

Related Resources

A separate real-recorded binaural dataset is planned/provided for real-domain evaluation and sim-to-real analysis. The simulated and real datasets are kept separate to distinguish controlled acoustic simulation data from real acoustic recordings.

License

This dataset is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0).

Users must provide appropriate attribution, use the dataset only for non-commercial purposes, and distribute derivative works under the same or a compatible license.

Downloads last month: 19