Customer-obsessed science

Automatically dubbed speech of a video involves: (i) segmenting the target sentences into phrases to reflect the speech-pause arrangement used by the original speaker, and (ii) adjusting the speaking rate of the synthetic voice at the phrase-level to match the exact timing of each corresponding source phrase. In this work, we investigate a post-segmentation approach to control the speaking rate of neural

Bayesian optimization (BO) is a popular method for optimizing expensive-to-evaluate black-box functions. BO budgets are typically given in iterations, which implicitly assumes each evaluation has the same cost. In fact, in many BO applications, evaluation costs vary significantly in different regions of the search space. In hyperparameter optimization, the time spent on neural network training increases

Many factors influence speech yielding different renditions of a given sentence. Generative models, such as variational autoencoders (VAEs), capture this variability and allow multiple renditions of the same sentence via sampling. The degree of prosodic variability depends heavily on the prior that is used when sampling. In this paper, we propose a novel method to compute an informative prior for the VAE

In real scenarios, a multilingual model trained to solve NLP tasks on a set of languages can be required to support new languages over time. Unfortunately, the straightforward retraining on a dataset containing annotated examples for all the languages is both expensive and time-consuming, especially when the number of considered languages grows. Moreover, the original annotated material may no longer be

Inspired by recent work in meta-learning and generative teaching networks, the authors propose a framework called Generative Conversational Networks, in which conversational agents learn to generate their own labelled training data (given some seed data) and then train themselves from that data to perform a given task.