In this work, we propose a novel method for modeling numerous speakers, which enables expressing the overall characteristics of speakers in detail like a trained multi-speaker model without additional training on the target speaker’s dataset. Although various works with similar purposes have been actively studied, their performance has not yet reached that of trained multi-speaker models due to their fundamental limitations. To overcome previous limitations, we propose effective methods for feature learning and representing target speakers’ speech characteristics by discretizing the features and conditioning them to a speech synthesis model. Our method obtained a significantly higher similarity mean opinion score (SMOS) in subjective similarity evaluation than seen speakers of a high-performance multi-speaker model, even with unseen speakers. The proposed method also outperforms a zero-shot method by significant margins. Furthermore, our method shows remarkable performance in generating new artificial speakers. In addition, we demonstrate that the encoded latent features are sufficiently informative to reconstruct an original speaker’s speech completely. It implies that our method can be used as a general methodology to encode and reconstruct speakers’ characteristics in various tasks.

In this paper, a method is proposed to measure human respiratory volume using a depth camera. The level-set segmentation method, combined with spatial and temporal information, was used to measure respiratory volume accurately. The shape of the human chest wall was used as spatial information. As temporal information, the segmentation result from the previous frame in the time-aligned depth image was used. The results of the proposed method were verified using a ventilator. The proposed method was also compared with other level-set methods. The result showed that the mean tidal volume error of the proposed method was 8.41% compared to the actual tidal volume. This was calculated to have less error than with two other methods: the level-set method with spatial information (14.34%) and the level-set method with temporal information (10.93%). The difference between these methods of tidal volume error was statistically significant [Formula: see text]. The intra-class correlation coefficient (ICC) of the respiratory volume waveform measured by a ventilator and by the proposed method was 0.893 on an average, while the ICC between the ventilator and the other methods were 0.837 and 0.879 on an average.