Breton is a minority language spoken in the Brittany region of France. Public initiatives are being undertaken in order to preserve the Breton language. As an effort toward that goal, we created a large Breton speech corpus and related automatic annotation tools. The corpus contains 20 hours of reading aloud for both a male and a female Breton speaker. Then, end-to-end text-to-speech synthesis systems are built. Subjective evaluation suggests that the systems are able to reproduce the voices of the original speakers faithfully. * David Guennec is now employed by ViaDialog * Hassan Hajipoor is now a PhD candidate at University of Massachusetts * Gwénolé Lecorvé is now employed by Orange Innovation

Effectiveness of websites is largely dependent on the quality of the website. The biggest share of the quality`s new concept is that the technical aspects of products and services combines with customers usage and understanding. Therefore websites evaluation based on the maximum usage and perception of the customers is considered an important issue to announce to the related organizations the success of website from customers' views. This customer relationship need a kind of management that first step of that for future decision needs knowledge about the websites features, customer insight and the position of websites among the competitors. One of the available media is the online news websites which their success is highly dependent on the relationship of their users. In this article achieving the information of websites is automatic and without the intervention of human so that the instant evaluation could be possible and used method is TOPSIS combined with information entropy to rank 791 news website which have most visitors of the Iranian users based on Alexa ranking report.

We propose an effective context-sensitive neural model for the task of time to event (TTE) prediction, which aims to predict the amount of time to/from the occurrence of given events in streaming content. We investigate this problem in the context of a multi-task learning framework, which we enrich with time difference embeddings. To conduct this research, we develop a multi-genre dataset of English events about soccer competitions and academy awards ceremonies, as well as their relevant tweets obtained from Twitter. Our model is 1.4 and 3.3 hours more accurate than the current state-of-the-art model in estimating TTE on English and Dutch tweets respectively. We examine different aspects of our model to illustrate its source of improvement. 1

Serial recall experiments study the ability of humans to recall words in the order in which they occurred. The following serial recall effects are generally investigated in studies with humans: word length and frequency, primacy and recency, semantic confusion, repetition, and transposition effects. In this research, we investigate LSTM language models in the context of these serial recall effects. Our work provides a framework to better understand and analyze neural language models and opens a new window to develop accurate language models.