Text-to-speech (TTS) aims to synthesize intelligible and natural speech based on the given text. It is a hot topic in language, speech, and machine learning research and has broad applications in industry. This book introduces neural network-based TTS in the era of deep learning, aiming to provide a good understanding of neural TTS, current research and applications, and the future research trend. This book first introduces the history of TTS technologies and overviews neural TTS, and provides preliminary knowledge on language and speech processing, neural networks and deep learning, and deep generative models. It then introduces neural TTS from the perspective of key components (text analyses, acoustic models, vocoders, and end-to-end models) and advanced topics (expressive and controllable, robust, model-efficient, and data-efficient TTS). It also points some future research directions and collects some resources related to TTS. This book is the first to introduce neural TTS in a comprehensive and easy-to-understand way and can serve both academic researchers and industry practitioners working on TTS.

What is the value of a social network? Prior work suggests two distinct mechanisms that have historically been difficult to differentiate: as a conduit of information, and as a source of social and economic support. We use a rich 'digital trace' dataset to link the migration decisions of millions of individuals to the topological structure of their social networks. We find that migrants systematically prefer 'interconnected' networks (where friends have common friends) to 'expansive' networks (where friends are well connected). A micro-founded model of network-based social capital helps explain this preference: migrants derive more utility from networks that are structured to facilitate social support than from networks that efficiently transmit information.

Nowadays, people receive many information through internet and have various channels to share these information. So it is important to understand agents' strategic information disclosure when they observe multiple signals. For instance, whether bidders in an oil-tract auction want to share some information about the oil tract or information about their own extracting technologies, whether experts being consulted by voters want to reveal or hide the information related to the alternatives in the vote, and how job hunters signal their skills by strategically choosing some tasks to perform. Information disclosure plays a key role, because it could change the nature of the game by changing both the information structure and agents' beliefs. Three popular applications are studied in this thesis. In a competition, agents may have incentives to reveal private information when there are both common and private components to their valuations and private information is held on both dimensions. When agents only observe one private signal, they do not have incentives to reveal, because revealing the only signal makes them fully lose their information advantage (suggested by literature). However, when agents observe multiple signals, they may have incentives to reveal some signals in order to earn higher profits from other signals. I show that there exists an equilibrium with full revelation of common-value signals and full concealment of private-value signals in competitions, including standard auctions and Bertrand oligopolies. This equilibrium achieves full efficiency and may give the seller a higher profit. Experiments confirm these theoretical predictions, such as subjects tend to hide private information in a pure common value environment, and choose to reveal the common-value signal in the presence of an additional private-value signal. In a voting, voters could consult a set of experts before voting over two alternatives. Agents have private biases over which alternative they prefer ex ante, but may be swayed by information about relative values of the alternatives. Experts observe private signals about the relative values of the alternatives and can choose to either reveal that information or conceal it, but they cannot lie. We examine how disclosure and voting behaviors vary with the intensity and heterogeneity of the preference biases, the informativeness of signals, and the structure of the voting rule. The voting rule that maximizes information disclosure can be a supermajority rule favoring one of the two alternatives, and in some cases unanimity rule can dominate simple majority rule both in terms of information revelation and total utility maximization, even in a fully symmetric society. The voting rule that maximizes information disclosure need not coincide with the voting rule that maximizes total utility. We also show that in a large enough society, full information revelation is approximated via any voting rule, and with sufficient symmetry simple majority rule is approximately ex ante efficient. In a job searching, multiple candidates (senders) compete over a fixed number of jobs. Senders' ability is unobservable, and determines their probability of success in different tasks. We study a signaling game with multiple senders each choosing one task to perform, and one receiver who observes all task choices and performances (success or failure) and matches senders to jobs. In order to analyze the effects of competition, we consider two refinements of the sequential equilibria: (i) sequential equilibria that survive when varying the number of senders; (ii) sequential equilibria that are supported by out-of-the-equilibrium-path beliefs satisfying a monotonicity condition (implied by Banks and Sobel's divinity refinement). We show that the set of sequential equilibria includes simple pooling equilibria where all senders choose the same task, and these simple pooling equilibria are the only type of sequential equilibria that satisfies (i). The unique sequential equilibrium under both (i) and (ii) is a simple pooling equilibrium with every sender choosing the most informative (difficult) task. In addition, if senders have a lower overall likelihood of success in more informative (difficult) tasks, this unraveling towards conspicuousness is inefficient. In summary, this thesis provides a new angle to several important applications where strategic information disclosure changes the nature of the games, the predictions of equilibrium behaviors and even the social welfare. Thus, considering information communicate becomes essential to both theoretical researches and policy suggestions. For instance, public announcements of verifiable evidences related to the common value should be encouraged in order to reduce information asymmetry and improve efficiency.

We examine the informal exchange of favors in societies such that any two individuals interact too infrequently to sustain exchange, but such that the social pressure of the possible loss of multiple relationships can sustain exchange. Patterns of exchange that are locally enforceable and renegotiation-proof necessitate that all links are "supported": any two individuals exchanging favors have a common friend. In symmetric settings, such robust networks are "social quilts": tree-like unions of completely connected subnetworks. Examining favor exchange networks in 75 villages in rural India, we find high levels of support and identify characteristics that correlate with support.

A set of voters consults experts before voting over two alternatives. Experts observe private signals about the values of the alternatives and can reveal their information or conceal it, but cannot lie. We examine how disclosure and voting vary with preference biases, signal precision, and the voting rule. Unanimity rule can lead to greater information revelation and total utility than simple majority rule. The voting rule that maximizes information disclosure need not coincide with the voting rule that maximizes total utility. In a large enough society, full information revelation is approximated via any voting rule.