Dimensionality reduced HRTFs: a comparative study

Dimensionality reduction is a statistical tool commonly used to map high-dimensional data into lower a dimensionality. The transformed data is typically more suitable for regression analysis or classification than the original data. Being of high dimensionality, HRTF data is commonly reduced using Principal Components Analysis (PCA). While highly effective at compressing data that follows the assumed model, PCA compression performance suffers when data follows a nonlinear distribution or when outliers are present. More recent data reduction techniques such as Isomap and locally linear embedding (LLE) take advantage of local neighborhood information in order to learn a more suitable basis for the HRTF data at hand. Quantitative results from previous work indicate that the embeddings created by both LLE and Isomap are superior to those obtained with PCA. This paper presents a study that compares sound source localization accuracy by human observers when presented with a virtual sound sy...

• Type: Conference
• Year: 2008
• Where: ACMACE
• Authors: Bill Kapralos, Nathan Mekuz, Agnieszka Kopinska, Saad Khattak
• Keywords: ACMACE 2008, Dimensionality Reduced Hrtfs, Dimensionality Reduction, HRTF Data, Human Computer Interaction