Kazuya Takeda

TL;DR: Experimental results demonstrate that the proposed sound event encoder is capable of converting sound events into their onomatopoeic representations with a 74.5% subjective acceptability rating, and that use of typical onomatography representations, as approved by multiple subjects, yields significant improvement, resulting in an acceptability rate of 81.8%.

TL;DR: A dialog strategy for information retrieval is regarded as a graph search problem and several novel dialog strategies that can recover from misrecognition through a spoken dialog that traverses the graph are proposed.

TL;DR: A multichannel speech enhancement method based on MAP speech spectral magnitude estimation using a generalized gamma model of speech prior distribution, where the model parameters are adapted from actual noisy speech in a frame-by-frame manner, resulting in better performance of speech enhancement algorithm.

TL;DR: In this article, a new loss function based on metric learning was proposed to learn the distance relationship from each class centroid in feature space for the binary classification model. But the performance of the proposed multi-task learning of the binary classifier and the metric learning makes it possible to build the feature space where the within class variance is minimized and the between-class variance is maximized while keeping normal classes linearly separable.

TL;DR: In this article, an evaluation framework for Japanese noisy speech recognition named AURORA-2J is introduced, which is a Japanese connected digits corpus and its evaluation scripts are designed in the same way as Aurora 2.