Previous works in zero-shot text-to-speech (ZS-TTS) have attempted to enhance its systems by enlarging the training data through crowd-sourcing or augmenting existing speech data. However, the use of low-quality data has led to a decline in the overall system performance. To avoid such degradation, instead of directly augmenting the input data, we propose a latent filling (LF) method that adopts simple but effective latent space data augmentation in the speaker embedding space of the ZS-TTS system. By incorporating a consistency loss, LF can be seamlessly integrated into existing ZS-TTS systems without the need for additional training stages. Experimental results show that LF significantly improves speaker similarity while preserving speech quality.

Adaptive text-to-speech (TTS) has attracted increasing interests for the purpose of training TTS systems without tons of high quality data. Nevertheless, existing adaptive TTS systems still show low adaptation quality for novel speakers, since it is hard to learn an extensive speaking style with limited data. To address this issue, we propose progressive variational autoencoder (PVAE) which generates data with adapting to style gradually. PVAE learns a progressively style-normalized representation, which is a key component of progressive style adaptation. We extend PVAE to PVAE-TTS, a multi-speaker adaptive TTS model which generates natural speech with high adaptation quality for novel speakers. To further improve the adaptation quality, we also propose dynamic style layer normalization (DSLN) which utilizes a convolution operation. The experimental results demonstrate the superiority of PVAE-TTS in terms of both subjective and objective evaluations.

This paper aims at building a knowledge-based system of smart mirrors for the cosmetic industry. In order to better understand hair and cosmetic experts, we first conduct interviews for knowledge acquisition. Then, we obtain insights from each category of answers collected from expert interviews. To design this knowledge-based system, we define concepts, main tasks and subtasks in order to extract rules to design the system. This system can suggest hairstyles and make-up colors by considering users’ face shapes and their personal colors.

This paper presents HierSpeech, a high-quality end-to-end text-to-speech (TTS) system based on a hierarchical conditional variational autoencoder (VAE) utilizing self-supervised speech representations. Recently, single-stage TTS systems, which directly generate raw speech waveform from text, have been getting interest thanks to their ability in generating high-quality audio within a fully end-to-end training pipeline. However, there is still a room for improvement in the conventional TTS systems. Since it is challenging to infer both the linguistic and acoustic attributes from the text directly, missing the details of attributes, specifically linguistic information, is inevitable, which results in mispronunciation and over-smoothing problem in their synthetic speech. To address the aforementioned problem, we leverage self-supervised speech representations as additional linguistic representations to bridge an information gap between text and speech. Then, the hierarchical conditional VAE is adopted to connect these representations and to learn each attribute hierarchically by improving the linguistic capability in latent representations. Compared with the state-of-the-art TTS system, HierSpeech achieves +0.303 comparative mean opinion score, and reduces the phoneme error rate of synthesized speech from 9.16% to 5.78% on the VCTK dataset. Furthermore, we extend our model to HierSpeech-U, an untranscribed text-to-speech system. Specifically, HierSpeech-U can adapt to a novel speaker by utilizing self-supervised speech representations without text transcripts. The experimental results reveal that our method outperforms publicly available TTS models, and show the effectiveness of speaker adaptation with untranscribed speech.

Although recent works on neural vocoder have improved the quality of synthesized audio, there still exists a gap between generated and ground-truth audio in frequency space. This difference leads to spectral artifacts such as hissing noise or reverberation, and thus degrades the sample quality. In this paper, we propose Fre-GAN which achieves frequency-consistent audio synthesis with highly improved generation quality. Specifically, we first present resolution-connected generator and resolution-wise discriminators, which help learn various scales of spectral distributions over multiple frequency bands. Additionally, to reproduce high-frequency components accurately, we leverage discrete wavelet transform in the discriminators. From our experiments, Fre-GAN achieves high-fidelity waveform generation with a gap of only 0.03 MOS compared to ground-truth audio while outperforming standard models in quality.

Few-shot speaker adaptation is a specific Text-to-Speech (TTS) system that aims to reproduce a novel speaker’s voice with a few training data. While numerous attempts have been made to the few-shot speaker adaptation system, there is still a gap in terms of speaker similarity to the target speaker depending on the amount of data. To bridge the gap, we propose GC-TTS which achieves high-quality speaker adaptation with significantly improved speaker similarity. Specifically, we leverage two geometric constraints to learn discriminative speaker representations. Here, a TTS model is pre-trained for base speakers with a sufficient amount of data, and then fine-tuned for novel speakers on a few minutes of data with two geometric constraints. Two geometric constraints enable the model to extract discriminative speaker embeddings from limited data, which leads to the synthesis of intelligible speech. We discuss and verify the effectiveness of GC-TTS by comparing it with popular and essential methods. The experimental results demonstrate that GC-TTS generates high-quality speech from only a few minutes of training data, outperforming standard techniques in terms of speaker similarity to the target speaker.

BACKGROUND: microRNAs (miRNAs) have been shown to play essential roles in a wide range of biological processes. Many computational methods have been developed to identify targets of miRNAs. However, the majority of these methods depend on pre-defined features that require considerable efforts and resources to compute and often prove suboptimal at predicting miRNA targets. RESULTS: We developed a novel hybrid deep learning-based (DL-based) approach that is capable of predicting miRNA targets at a higher accuracy. This approach integrates convolutional neural networks (CNNs) that excel in learning spatial features and recurrent neural networks (RNNs) that discern sequential features. Therefore, our approach has the advantages of learning both the intrinsic spatial and sequential features of miRNA:target. The inputs for our approach are raw sequences of miRNAs and genes that can be obtained effortlessly. We applied our approach on two human datasets from recently miRNA target prediction studies and trained two models. We demonstrated that the two models consistently outperform the previous methods according to evaluation metrics on test datasets. Comparing our approach with currently available alternatives on independent datasets shows that our approach delivers substantial improvements in performance. We also show with multiple evidences that our approach is more robust than other methods on small datasets. Our study is the first study to perform comparisons across multiple existing DL-based approaches on miRNA target prediction. Furthermore, we examined the contribution of a Max pooling layer in between the CNN and RNN and demonstrated that it improves the performance of all our models. Finally, a unified model was developed that is robust on fitting different input datasets. CONCLUSIONS: We present a new DL-based approach for predicting miRNA targets and demonstrate that our approach outperforms the current alternatives. We supplied an easy-to-use tool, miTAR, at https://github.com/tjgu/miTAR . Furthermore, our analysis results support that Max Pooling generally benefits the hybrid models and potentially prevents overfitting for hybrid models.

This study examined which accompanying information elements were noticed by visitors while they were looking at artworks, using eye-tracking experiments. First, we conducted an online survey to grasp the types of information that visitors wanted to know, and five elements were obtained. Second, we collected information on these five elements through interviews with one artist. Third, eye-tracking experiments were performed with semi-structured interviews. We set the information delivery media as follows: wall and mobile texts as commonly used in art museums. The results showed that patterns of eye movement of visitors were different according to the information delivery media. Also, we found that there was a correlation between the results of the eye-tracking experiment and visitor interest. This study has limitations in that it is an experiment limited to small sample size and artwork genre; however, it is meaningful in that it was able to grasp the information needs of visitors through eye-tracking.

To enhance missing nonverbal cues in computer-mediated communication using text, those who can see often use emojis or emoticons. Although emojis for the sighted have transformed throughout the years to animated forms and added sound effects, emojis for visually impaired people remain underdeveloped. This study tested how tactile emojis based on visual imagery combined with the Braille system can enhance clarity in the computer-mediated communication environment for those with visual impairments. Results of this study confirmed three things: Visually impaired subjects were able to connect emotional emojis to the emotion they represented without any prior guidance, image-based (picture-based) and non-image-based (abstraction-based) tactile emoji were equally learnable, and the clarity of intended meaning was improved when an emoji was used with text (Braille). Thirty visually impaired subjects were able to match an average of 67% of emotions without prior guidance, and three of the four subjects who matched perfectly both before and after guidance were congenitally blind. The subjects had the most trouble discriminating the facial feature of “fear” between “sadness” or “surprised” for they shared similar traits. After guidance, the image-based tactile design elicited an average of 81% correct answers, whereas the non-image-based tactile design elicited an average of 37%, showing that the image-based tactile design was more effective for learning the meaning of emojis. The clarity of the sentence was also improved. This study shows that image-based tactile emojis can improve the texting experience of visually impaired individuals to a level where they can communicate subtle emotional cues through tactile imagery. This advance could minimize the service gap between sighted and visually impaired people and offer a much more abundant computer-mediated communication environment for visually impaired individuals.

SUMMARY: Immune diseases have a strong genetic component with Mendelian patterns of inheritance. While the tight association has been a major understanding in the underlying pathophysiology for the category of immune diseases, the common features of these diseases remain unclear. Based on the potential commonality among immune genes, we design Gene Ranker for key gene identification. Gene Ranker is a network-based gene scoring algorithm that initially constructs a backbone network based on protein interactions. Patient gene expression networks are added into the network. An add-on process screens the networks of weighted gene co-expression network analysis (WGCNA) on the samples of immune patients. Gene Ranker is disease-specific; however, any WGCNA network that passes the screening procedure can be added on. With the constructed network, it employs the semi-supervised learning for gene scoring. RESULTS: The proposed method was applied to immune diseases. Based on the resulting scores, Gene Ranker identified potential key genes in immune diseases. In scoring validation, an average area under the receiver operating characteristic curve of 0.82 was achieved, which is a significant increase from the reference average of 0.76. Highly ranked genes were verified through retrieval and review of 27 million PubMed literatures. As a typical case, 20 potential key genes in rheumatoid arthritis were identified: 10 were de facto genes and the remaining were novel. AVAILABILITY AND IMPLEMENTATION: Gene Ranker is available at http://www.alphaminers.net/GeneRanker/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Small P-values are often required to be accurately estimated in large-scale genomic studies for the adjustment of multiple hypothesis tests and the ranking of genomic features based on their statistical significance. For those complicated test statistics whose cumulative distribution functions are analytically intractable, existing methods usually do not work well with small P-values due to lack of accuracy or computational restrictions. We propose a general approach for accurately and efficiently estimating small P-values for a broad range of complicated test statistics based on the principle of the cross-entropy method and Markov chain Monte Carlo sampling techniques. RESULTS: We evaluate the performance of the proposed algorithm through simulations and demonstrate its application to three real-world examples in genomic studies. The results show that our approach can accurately evaluate small to extremely small P-values (e.g. 10-6 to 10-100). The proposed algorithm is helpful for the improvement of some existing test procedures and the development of new test procedures in genomic studies. AVAILABILITY AND IMPLEMENTATION: R programs for implementing the algorithm and reproducing the results are available at: https://github.com/shilab2017/MCMC-CE-codes. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Gene set analysis is a powerful method of deducing biological meaning for an a priori defined set of genes. Numerous tools have been developed to test statistical enrichment or depletion in specific pathways or gene ontology (GO) terms. Major difficulties towards biological interpretation are integrating diverse types of annotation categories and exploring the relationships between annotation terms of similar information. RESULTS: GARNET (Gene Annotation Relationship NEtwork Tools) is an integrative platform for gene set analysis with many novel features. It includes tools for retrieval of genes from annotation database, statistical analysis & visualization of annotation relationships, and managing gene sets. In an effort to allow access to a full spectrum of amassed biological knowledge, we have integrated a variety of annotation data that include the GO, domain, disease, drug, chromosomal location, and custom-defined annotations. Diverse types of molecular networks (pathways, transcription and microRNA regulations, protein-protein interaction) are also included. The pair-wise relationship between annotation gene sets was calculated using kappa statistics. GARNET consists of three modules--gene set manager, gene set analysis and gene set retrieval, which are tightly integrated to provide virtually automatic analysis for gene sets. A dedicated viewer for annotation network has been developed to facilitate exploration of the related annotations. CONCLUSIONS: GARNET (gene annotation relationship network tools) is an integrative platform for diverse types of gene set analysis, where complex relationships among gene annotations can be easily explored with an intuitive network visualization tool (http://garnet.isysbio.org/ or http://ercsb.ewha.ac.kr/garnet/).

A simple contract contains formalizing preconditions, postconditions, and invariants. A contract describes the services that are provided by an object. In the component world, it is essential to describe correctly what services are provided by the components and how we can use them. Till now, interfaces are specified in IDL, which describes the syntactic aspect of interface. IDL cannot describe the semantic aspects and safety condition of a component that a client and a server must know how to use or implement it. In this paper, we use contracts specifying components and propose an interface specification technique to describe components as a contract(namely, component contract). We regard a component's interface as contract and present which aspect to be specified for providing correct information of the components. This paper defines some definitions and describes operators such as cooperating components, component version, functional requirements, nonfunctional requirements, and performance measurements to specify components, and specifies them using Z scheme. Finally, we apply this interface specification technique to develop an e-commerce system.