1,721,036 research outputs found
CALCOLO DELLO SPEECH RATE MEDIANTE UN ALGORITMO DI RICERCA DI NUCLEI SILLABICI IN BASE AL PROFILO DELL'ENERGIA
No full textThough this be hesitant, yet there is method in ’t: Effects of disfluency patterns in neural speech synthesis for cultural heritage presentations
No full textAutomatic classification of emotions via global and local prosodic features on a multilingual emotional database
No full text
Overview of the evalita 2018 spoken utterances guiding chef’s assistant robots (SUGAR) task
No full textThe SUGAR task is intended to develop a baseline to train a voicecontrolled robotic agent to act as a cooking assistant. The starting point will be therefore to provide authentic spoken data collected in a simulated natural context from which semantic predicates will be extracted to classify the actions to perform. Three different approaches were used by the two SUGAR participants to solve the task. The enlightening results show the different elements of criticality underlying the task itself
Conflict Search Graph for Common Ground Consistency checks in Dialogue Systems
No full textIn this work, we account for the formalisation of a Conflict Search Graph as a module managing domain knowledge, dialogue state tracking and information consistency in dialogue systems. Insights on its ability to recognise Common Ground Inconsistencies and make them explicit via specific linguistic feedback are also reported
On the Impact of Location-related Terms in Neural Embeddings for Content Similarity Measures in Cultural Heritage Recommender Systems
No full textAnalysing text to detect semantic similarities is a recent breakthrough of Natural Language Processing that brought many novel applications in different fields. A domain which could greatly benefit of this innovation is the one regarding Location-based and/or Touristic Recommender Systems, where the user receives suggestions based on his/her past liked items. In this work, we consider the use of neural embeddings weighted using Smooth-Inverse Frequency (SIF) to detect semantic similarities in textual descriptions found in a large graph database covering Italian cultural Points of Interests (POIs). Of all detected similar pairs on a national scale, 19% are composed by POIs that do not belong to the same ontological category, highlighting the potential neural embeddings have to match POIs beyond the categories they have been assigned to. However, since text descriptions also contain references to the places where POIs are found, similarities can be detected among POIs sharing the same location, especially in the case of low-frequency geographical terms. While this may be desirable, in some cases, it may harm location-aware applications, as POIs positions are already known. By comparing city names occurrence probabilities both in the full text corpus and in location-constrained sub-corpora, we observed probability shifts, on average, of 232%. This suggests that, for the specific case of location-aware services, SIF-weighted neural embeddings should use location-constrained sub-corpora for term occurrence probability computation in order to efficiently remove uninteresting information
On-Line Filtering of On-Street Parking Data to Improve Availability Predictions
No full textKnowing where to park in advance is a most wished feature by many drivers. In recent years, many research efforts have been spent to analyse massive amount of parking information, to learn availability trends and thus to predict, within a Parking Guidance and Information (PGI) system, where there is the highest chance to find free parking spaces. The most of these solutions exploits raw data coming from stationary sensors or crowd-sensed by mobile probes. In both the cases, these massive amounts of data present a high level of noise, which heavily affects the quality of availability predictions. In a previous work we demonstrated that a 2-step approach, based on machine learning techniques to filter out noise, improves the quality of parking availability predictions over raw data. In this paper we propose a further advancement of that approach, by including a technique to perform such noise filtering in real-time, with reduced computational efforts. The proposal has been empirically tested on a real-world dataset of on-street parking information from the SFpark project, and compared against a regression model based on SVR, to perform parking availability predictions. Results show that the predictions obtained with the new on-line approach show a better balance between average and entropy in errors distribution with respect to the use of raw data coming from the sensors
Syllable classification using static matrices and prosodic features
Get PDFIn this paper we explore the usefulness of prosodic features for
syllable classification. In order to do this, we represent the
syllable as a static analysis unit such that its acoustic-temporal
dynamics could be merged into a set of features that the SVM
classifier will consider as a whole. In the first part of our
experiment we used MFCC as features for classification,
obtaining a maximum accuracy of 86.66%. The second part of
our study tests whether the prosodic information is
complementary to the cepstral information for syllable
classification. The results obtained show that combining the
two types of information does improve the classification, but
further analysis is necessary for a more successful
combination of the two types of features
- …
