A quick search method coping with multiple kinds of feature distortions

摘要

Here we propose a quick search method for distorted audio signals. The method is an application of the quick OR-search method that we previously developed. In the present method, feature distortions are first learned as probabilistic distributions, and then the multiple distributions are efficiently considered using the OR-search method. The experiments using microphone-recorded signals show that the proposed method achieves approximately 3 to 5% better accuracy than the old method with little increase of search time.