Naver Corporation

About: Naver Corporation is a company organization based out in Seongnam-si, South Korea. It is known for research contribution in the topics: Terminal (electronics) & Computer science. The organization has 4038 authors who have published 4294 publications receiving 35045 citations. The organization is also known as: NAVER Corporation & NAVER.

Short utterance compensation in speaker verification via cosine-based teacher-student learning of speaker embeddings

Abstract: The short duration of an input utterance is one of the most critical threats that degrade the performance of speaker verification systems. This study aimed to develop an integrated text-independent speaker verification system that inputs utterances with short duration of 2 seconds or less. We propose an approach using a teacher-student learning framework for this goal, applied to short utterance compensation for the first time in our knowledge. The core concept of the proposed system is to conduct the compensation throughout the network that extracts the speaker embedding, mainly in phonetic-level, rather than compensating via a separate system after extracting the speaker embedding. In the proposed architecture, phonetic-level features where each feature represents a segment of 130 ms are extracted using convolutional layers. A layer of gated recurrent units extracts an utterance-level feature using phonetic-level features. The proposed approach also adopts a new objective function for teacher-student learning that considers both Kullback-Leibler divergence of output layers and cosine distance of speaker embeddings layers. Experiments were conducted using deep neural networks that take raw waveforms as input, and output speaker embeddings on VoxCeleb1 dataset. The proposed model could compensate approximately 65 \% of the performance degradation due to the shortened duration.

Autoencoder-based personalized ranking framework unifying explicit and implicit feedback for accurate top-N recommendation

Abstract: Existing top- N recommendation models can be classified according to the following two criteria: way of optimization and type of data. In terms of optimization, the models can either minimize the mean squared error (MSE) of rating predictions, which is so-called pointwise learning, or maximize the likelihood of pairwise preferences over more preferred and less preferred items (e.g., rated and unrated items), which is so-called pairwise learning. According to the data type, the models use either explicit feedback or implicit feedback. Most existing models use one of the optimization methods with either explicit or implicit feedback. However, we believe that pairwise learning and pointwise learning (resp. using explicit and implicit feedback) are complementary, thus employing both optimization methods and both forms of data together would bring a synergy effect in recommendation. Along this line, we propose a novel, unified recommendation framework based on deep neural networks, in which the pointwise and pairwise learning are employed together while using both the users’ explicit and implicit feedback. The experimental results on four real-life datasets confirm the effectiveness of our proposed framework over the state-of-the-art ones.

LP-WaveNet: Linear Prediction-based WaveNet Speech Synthesis

Abstract: We propose a linear prediction (LP)-based waveform generation method via WaveNet vocoding framework. A WaveNet-based neural vocoder has significantly improved the quality of parametric text-to-speech (TTS) systems. However, it is challenging to effectively train the neural vocoder when the target database contains massive amount of acoustical information such as prosody, style or expressiveness. As a solution, the approaches that only generate the vocal source component by a neural vocoder have been proposed. However, they tend to generate synthetic noise because the vocal source component is independently handled without considering the entire speech production process; where it is inevitable to come up with a mismatch between vocal source and vocal tract filter. To address this problem, we propose an LP-WaveNet vocoder, where the complicated interactions between vocal source and vocal tract components are jointly trained within a mixture density network-based WaveNet model. The experimental results verify that the proposed system outperforms the conventional WaveNet vocoders both objectively and subjectively. In particular, the proposed method achieves 4.47 MOS within the TTS framework.

성별에 따른 한국 청소년의 건강행태, 정신건강과 인터넷 중독과의 관련성

Abstract: The purpose of this study was to compare the health behaviors, mental health and internet addiction by gender differences among Korean adolescents and to examine relevances between health behaviors, mental health and internet addiction The subjects were 73,238 Korean adolescents(male: 38,391, female: 34,847) who were recruited through national web-based survey The data were derived from the Sixth Korea Youth Risk Behavior Web-based Survey 2010 in South Korea There were significant differences in health behaviors, mental health and internet addiction by gender differences Prevalence of internet addiction was male: 39%, female: 19% in this study As the result of multiple logistic regression, the risk of internet addiction was increased in the case of smoking experience, substance experience, subjective health status, feeling of stress, depression experience, suicidal ideation, feeling of happiness, and sufficiency of sleeps in both male and female Therefore, to decrease internet addiction prevalence, it is necessary to detect and manage the influencing risk factors of internet addiction such as health behaviors and mental health And the health-promotion intervention to improve the internet addiction of adolescents should be planed and provided considering such differences by gender differences

Data Refinement and Channel Selection for a Portable E-Nose System by the Use of Feature Feedback

Abstract: We propose a data refinement and channel selection method for vapor classification in a portable e-nose system. For the robust e-nose system in a real environment, we propose to reduce the noise in the data measured by sensor arrays and distinguish the important part in the data by the use of feature feedback. Experimental results on different volatile organic compounds data show that the proposed data refinement method gives good clustering for different classes and improves the classification performance. Also, we design a new sensor array that consists only of the useful channels. For this purpose, each channel is evaluated by measuring its discriminative power based on the feature mask used in the data refinement. Through the experimental results, we show that the new sensor array improves both the classification rates and the efficiency in computation and data storage.