/library/oar/handle/123456789/52654 2025-12-26T16:01:52Z 2025-12-26T16:01:52Z /library/oar/handle/123456789/74891 2021-04-29T05:02:31Z 2019-01-01T00:00:00Z

Title: A diphone-based Maltese speech synthesis system Abstract: While there has been work in the area, at the time of writing there are no available TTS systems for Maltese, thus almost the entire system had to be built from scratch. In light of this, a Diphone-Based Concatenative Speech System was chosen as the type of synthesiser to implement. This was due to the minimal amount of data needed, requiring less than 20 minutes of recorded speech. A simple `Text Normalisation' component was built, which converts integers between 0 and 9,999 written as numerals to their textual form. While this is far from covering all the possible forms of Non-Standard Words (NSWs) in Maltese, the modular nature in which it was built allows for easy upgrading in future work. A `Grapheme to Phoneme (G2P)' component which then converts the normalised text into a sequence of phonemes (basic sounds) that make up the text was also created, based on an already existing implementation by Crimsonwing. Three separate `Diphone Databases' were made available to the speech synthesiser. One of these is the professionally recorded English Diphone database FestVox's `CMU US KAL Diphone'1. The second and third were created as part of this work, one with diphones manually extracted from the recorded carrier phrases in Maltese, the other with diphones automatically extracted using Dynamic Time Warping (DTW). The Time Domain - Pitch Synchronous OverLap Add (TD-PSOLA) concatenation algorithm was implemented to string together the diphones in the sequence specified by the G2P component. On a scale of 1 to 5, the speech synthesised when using the diphone database of manually extracted diphones concatenated by the TD-PSOLA algorithm was scored 2.57 for naturalness, 2.72 for clarity, and most important of all, 3.06 for Intelligibility by evaluators. These scores were higher than those obtained when using the professionally recorded English diphone set. Description: B.SC.ICT(HONS)ARTIFICIAL INTELLIGENCE

2019-01-01T00:00:00Z /library/oar/handle/123456789/74889 2021-04-29T05:01:49Z 2019-01-01T00:00:00Z

Title: Mining drug-drug interactions for healthcare professionals Abstract: The fourth leading cause of death in the US are Adverse Drug Reactions (ADRs)red. One such cause of ADRs is brought about through Drug-drug Interactions (DDIs). The positive side of this is that such reactions can be prevented. DDIs are reported during the pharmacovigilance (PV) process. PV is the practice of monitoring and detecting ADRs once a drug is launched into the market. ¸£ÀûÔÚÏßÃâ·Ñ related to DDIs is dispersed across different biomedical articles. We propose medicX, a system that is able to detect DDIs in biomedical texts by leveraging on different machine learning techniques. The main components within our system are the Drug Named Entity Recognition (DNER) component and the DDI Identification component. Different approaches were investigated in line with existing research. The DNER component is evaluated using the CHEMDNER and the DDIExtraction 2013 challenge corpora. Conversely, the DDI Identification component is evaluated using the DDIExtraction 2013 challenge corpus. The DNER component is implemented using an approach based on LSTM-CRF. This method achieves a macro-averaged F1-score of 84.89% when it is trained and evaluated on the DDI-2013 corpus, which is 1.43% higher than the system that placed first in the DDIExtraction 2013 challenge. On the other hand, the DDI Identification component is implemented using a two-stage rich feature-based linear-kernel SVM. This classifier achieves an F1-score of 66.18%, as compared to the SVM state-of-the-art DDI system that reported an F1-score of 71.79%. Description: B.SC.ICT(HONS)ARTIFICIAL INTELLIGENCE

2019-01-01T00:00:00Z /library/oar/handle/123456789/74875 2021-04-29T05:00:55Z 2019-01-01T00:00:00Z

Title: dOMiNiuM PubliCuM : opinion mining of news portal comments Abstract: Sentiment analysis is a research problem with great potential, given the enormous applications of being able to accurately summarise the opinion expressed by a person towards any topic. It has seen a lot of research into product reviews over the years, unfortunately research into sentiment analysis on more ambiguous data like news portal comments has been far more limited due to greater challenges. We propose a rules-based aspect level sentiment analysis using an opinion word corpus to detect the sentiment expressed towards entities in user-generated content, noted to be one of the more complex forms of data. The system is designed to use comments extracted from the Times of Malta news portal. Following the extraction of comments each sentence is processed to identify if it is in English or not to ensure that only English sentences are processed further. Sentences not in the English language are simply marked and stored. The English sentences within a comment are each processed for sentiment analysis. The scores from each sentence contribute to the sentiment mean value of each entity in the specific comment. Experimental results indicate that this approach looks promising for similar endeavours in the future. Description: B.SC.ICT(HONS)ARTIFICIAL INTELLIGENCE

2019-01-01T00:00:00Z /library/oar/handle/123456789/74844 2021-04-28T05:12:14Z 2019-01-01T00:00:00Z

Title: A text-independent, multi-lingual and cross-corpus evaluation of emotion recognition in speech Abstract: Ongoing research on Human Computer Interaction (HCI) is always progressing and the need for machines to detect human emotion continues to increase for the purposes of having more personalized systems which can intelligently act according to user emotion. Varying languages may portray emotions di↵erently which is a hiccup in the field of automatic emotion recognition from speech. We propose a system which takes a cross-corpus and multilingual approach to emotion recognition from speech in order to show the behaviour of results when compared to single monolingual corpus testing. We utilize four di↵erent classifiers: K-Nearest Neighbours (KNN), Support Vector Machines (SVM), Multi-Layer Perceptrons (MLP), Gaussian Mixture Models (GMM) along with two di↵erent feature sets including Mel-Frequency Cepstral Coefficients (MFCCs) and our own extracted prosodic feature set on three di↵erent emotional speech corpora containing of several languages. The aim for the prosodic feature set is to try and acquire a general feature set that works well across all languages and corpora. We notice a drop in performance when unseen data is tested but made better when merged databases are present in the training data and when EMOVO is present in either training or testing. MFCCs work very well with GMMs on single corpus testing but our prosodic feature set works better in general on the rest of the classifiers. We evaluate all the obtained results in view of proving any elements that could possibly form a language independent system but for the time being results show otherwise. Description: B.SC.ICT(HONS)ARTIFICIAL INTELLIGENCE

2019-01-01T00:00:00Z