No abstract

No abstract

For building flexible and appealing high-quality speech synthesisers, it is desirable to be able to accommodate and reproduce fine variations in vocal expression present in natural speech. Synthesisers can enable control over such output properties by adding adjustable control parameters in parallel to their text input. If not annotated in training data, the values of these control inputs can be optimised jointly with the model parameters. We describe how this established method can be seen as approximate maximum likelihood and MAP inference in a latent variable model. This puts previous ideas of (learned) synthesiser inputs such as sentence-level control vectors on a more solid theoretical footing. We furthermore extend the method by restricting the latent variables to orthogonal subspaces via a sparse prior. This enables us to learn dimensions of variation present also within classes in coarsely annotated speech. As an example, we train an LSTM-based TTS system to learn nuances in emotional expression from a speech database annotated with seven different acted emotions. Listening tests show that our proposal successfully can synthesise speech with discernible differences in expression within each emotion, without compromising the recognisability of synthesised emotions compared to an identical system without learned nuances.

Preferentially processing behaviorally relevant information is vital for primate survival. In visuospatial attention studies, manipulating the spatial extent of attention focus is an important question. Although many studies have claimed to successfully adjust attention field size by either varying the uncertainty about the target location (spatial uncertainty) or adjusting the size of the cue orienting the attention focus, no systematic studies have assessed and compared the effectiveness of these methods. We used a multiple cue paradigm with 2.5° and 7.5° rings centered around a target position to measure the cue size effect, while the spatial uncertainty levels were manipulated by changing the number of cueing positions. We found that spatial uncertainty had a significant impact on reaction time during target detection, while the cue size effect was less robust. We also carefully varied the spatial scope of potential target locations within a small or large region and found that this amount of variation in spatial uncertainty can also significantly influence target detection speed. Our results indicate that adjusting spatial uncertainty is more effective than varying cue size when manipulating attention field size.

Record linkage is the task of identifying which records in one or more data collections refer to the same entity, and address is one of the most commonly used fields in databases. Hence, segmentation of the raw addresses into a set of semantic fields is the primary step in this task. In this paper, we present a probabilistic address parsing system based on the Hidden Markov Model. We also introduce several novel approaches of synthetic training data generation to build robust models for noisy real-world addresses, obtaining 95.6% F-measure. Furthermore, we demonstrate the viability and efficiency of this system for large-scale data by scaling it up to parse billions of addresses.

In the case that the background scene is dense map regularization complex and the detected objects are low texture, the method of matching according to the feature points is not applicable. Usually, the template matching method is used. When training samples are insufficient, the template matching method gets a worse detection result. In order to resolve the problem stably in real time, we propose a fast template matching algorithm based on the principal orientation difference feature. The algorithm firstly obtains the edge direction information by comparing the images that are binary. Then, the template area is divided where the different features are extracted. Finally, the matching positions are searched around the template. Experiments on the videos whose speed is 30 frames/s show that our algorithm detects the lowtexture objects in real time with a matching rate of 95%. Compared with other state-of-art methods, our proposed method reduces the training samples significantly and is more robust to the illumination changes.

No abstract

No abstract