13:20 - 2:20 (Posters and Demos)
Yichi Zhang, Zhiyao Duan
University of Rochester
Sound Search by Vocal Imitation
Searching sounds with text labels is often problematic and time consuming, as text labels do not often describe the detailed audio content. Query by example is a way to improve the effectiveness and efficiency of sound retrieval. We propose a novel approach for sound query by example: query by vocal imitation. Vocal imitation is commonly used in human communication and can be employed for novel human-computer interaction. We propose two related systems. The supervised system addresses the retrieval problem by vocal imitation recognition. It trains a multi-class classifier using training vocal imitations of different sound classes in the library and classifies a new imitation query into one of the classes. This system thus cannot retrieve sounds that are not trained. The unsupervised system is more flexible that it measures the feature distance between the imitation query and each sound in the library and returns sounds the most similar to the query.
Design downloaded from free website templates.