Article
Acoustics
Jesus Monge Alvarez, Holly Francois, Hosang Sung, Seungdo Choi, Jonghoon Jeong, Kihyun Choo, Kyoungbo Min, Sangjun Park
Summary: Spoken language is crucial for human-machine interaction, and text-to-speech (TTS) models are essential for efficient communication. CAMNet, based on deep convolutional TTS (DCTTS), offers controllable style transfer capabilities and allows for consistent control over expression, pitch, and speaking rate, while maintaining high-quality synthesized speech.
Article
Automation & Control Systems
Jieyu An, Wan Mohd Nazmee Wan Zainon
Summary: Multimodal sentiment analysis is an important research area, especially in social media where emotions are expressed through text and images. This paper proposes a novel model called ICCI, which integrates color cues to improve sentiment analysis accuracy. The model extracts semantic and color features, and utilizes a cross-attention mechanism for feature interaction. Experimental results on benchmark datasets demonstrate the effectiveness of ICCI, outperforming existing methods with higher accuracy.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2023)
Article
Chemistry, Multidisciplinary
Zengqiang Shang, Peiyang Shi, Pengyuan Zhang, Li Wang, Guangying Zhao
Summary: We propose a highly expressive end-to-end text-to-waveform generation model, which deeply couples the hierarchical properties of speech with hierarchical variational auto-encoders and models multi-scale latent variables. Our model performs closer to natural speech in prosody expressiveness and has better generative diversity.
APPLIED SCIENCES-BASEL
(2023)
Article
Chemistry, Multidisciplinary
Su Yang, Farzin Deravi
Summary: This paper proposes a novel re-engineering mechanism for generating word embeddings to enhance document-level sentiment analysis. By re-engineering the feature components of embedding vectors, the mechanism increases the between-class separation and leverages the informative content of the documents.
APPLIED SCIENCES-BASEL
(2022)
Article
Computer Science, Artificial Intelligence
Jose Antonio Garcia-Diaz, Salud Maria Jimenez-Zafra, Miguel Angel Garcia-Cumbreras, Rafael Valencia-Garcia
Summary: The rise of social networks has allowed individuals with misogynistic, xenophobic, and homophobic views to spread hate-speech, causing harm to individuals or groups based on their gender, ethnicity, or sexual orientation. Automatic identification of hate-speech is challenging, especially in languages other than English. This study focuses on identifying hate-speech in Spanish and examines the most effective features and their combination for developing accurate systems.
COMPLEX & INTELLIGENT SYSTEMS
(2023)
Article
Computer Science, Information Systems
Saadin Oyucu
Summary: The study developed a Turkish speech synthesis system using a deep learning approach to address the lack of corpus for Turkish TTS. Real users rated the quality of synthesized speech as 4.49 using Mean Opinion Score (MOS), and an objective evaluation obtained a score of 4.32. These findings represent the first documented deep learning and HiFi-GAN-based TTS system for Turkish TTS.
Article
Chemistry, Multidisciplinary
Cici Suhaeni, Hwan-Seung Yong
Summary: This paper examines the effectiveness of the GPT-3 model in addressing imbalanced sentiment analysis, specifically focusing on the imbalanced Coursera online course review dataset. The study employs synthetic review generation and sentiment classification using nine models on both imbalanced and balanced datasets. The results show that high-quality synthetic reviews significantly enhance sentiment classification performance, with an average accuracy increase of approximately 12.76% on the balanced dataset. The study highlights the potential of the GPT-3 model as a feasible solution for data imbalance in sentiment analysis and provides significant insights for future research.
APPLIED SCIENCES-BASEL
(2023)
Article
Computer Science, Information Systems
Arwa Alshehri, Abdulmohsen Algarni
Summary: In text classification tasks, feature representation and weighting schemes are important for classification performance. Traditional unsupervised term weighting (UTW) schemes, such as TF-IDF, are not sufficient for sentiment analysis (SA) tasks. This study proposes a novel supervised term weighting (STW) approach called TF-TDA, which categorizes extracted features into groups with different levels of discrimination and weights each group based on its contribution. Experimental results using four SA datasets show that TF-TDA outperforms two baseline term weighting approaches, with improvements in the F1 score ranging from 0.52% to 3.99%. Statistical tests confirm the significant improvement achieved by TF-TDA, with p-values ranging from 0.0000597 to 0.0455.
Review
Computer Science, Artificial Intelligence
Ganesh Chandrasekaran, Tu N. Nguyen, D. Jude Hemanth
Summary: Sentiment analysis is crucial for identifying and classifying opinions on products or services, with traditional text-based methods no longer meeting the needs of analyzing multimodal data effectively.
WILEY INTERDISCIPLINARY REVIEWS-DATA MINING AND KNOWLEDGE DISCOVERY
(2021)
Review
Computer Science, Artificial Intelligence
M. P. Jasir, Kannan Balakrishnan
Summary: This article discusses the research progress of text-to-speech synthesis in English and prominent Indian languages, with a special focus on Malayalam. It emphasizes the importance of improving the naturalness of synthetic speech in multilingual countries.
ACM TRANSACTIONS ON ASIAN AND LOW-RESOURCE LANGUAGE INFORMATION PROCESSING
(2022)
Article
Computer Science, Artificial Intelligence
Osman Semih Albayrak, Tevfik Aytekin, Tolga Ahmet Kalayci
Summary: Having a clean and standardized product catalog is crucial for e-commerce companies. This study introduces a novel duplicate record detection engine developed for an e-commerce company, Hepsiburada, and demonstrates its high precision in detecting duplicate product records.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Computer Science, Information Systems
P. Kumaran, Rajeswari Sridhar, Hiran Nandy
Summary: Text Sentiment Analysis (TSA) for blogs on major microblogging platforms is important but challenging due to the complexity of natural language and the informal structure employed in short text like Twitter. In this proposed work, a MLP-SDLM model is used to concatenate data filtering and feature engineering approaches, and a K-map based technique is introduced to efficiently combine filtered and unfiltered textual and non-textual features. The proposed models outperform traditional machine learning and deep learning classifiers, achieving high accuracy rates of 95.13% for MLP-SDLM, 89.17% for K-map based technique, and 88.7% for MLP.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Sarah A. Abdu, Ahmed H. Yousef, Ashraf Salem
Summary: This research provides a comprehensive overview of the latest updates in the field of video sentiment analysis, categorizing thirty-five state-of-the-art models based on the architecture used in each model. It concludes that the most powerful architecture in multimodal sentiment analysis task is the Multi-Modal Multi-Utterance based architecture.
INFORMATION FUSION
(2021)
Article
Computer Science, Artificial Intelligence
Xiaofei Zhu, Zhanwang Peng, Jiafeng Guo, Stefan Dietze
Summary: Sentiment classification aims to predict the sentiment label for a given text. Recent research efforts have focused on incorporating matching clues between text words and class labels into the learning process. However, these methods heavily rely on label content availability and only capture label-specific signals to measure word contribution. In this paper, a novel framework called LGDSC is proposed, which generates an effective label description and utilizes a Dual-Channel Label-guided Attention Network (DLAN) to learn text representation from two different channels.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Review
Chemistry, Multidisciplinary
Yili Wang, Jiaxuan Guo, Chengsheng Yuan, Baozhu Li
Summary: Twitter Sentiment Analysis is an active subfield of text mining, which has attracted considerable interest among researchers. This research provides a comprehensive review of the latest developments in this area, including newly proposed algorithms and applications. The survey classifies each publication based on its significance to specific TSA methods and depicts the current research direction in the field of TSA.
APPLIED SCIENCES-BASEL
(2022)
Article
Chemistry, Analytical
Rosa Ma Alsina-Pages, Ferran Orga, Francesc Alias, Joan Claudi Socoro
Review
Engineering, Electrical & Electronic
Francesc Alias, Rosa Ma. Alsina-Pages
JOURNAL OF SENSORS
(2019)
Article
Acoustics
Rosa Ma Alsina-Pages, Francesc Alias, Joan Claudi Socoro, Ferran Orga, Roberto Benocci, Giovanni Zambon
Article
Chemistry, Multidisciplinary
Marc Freixes, Marc Arnela, Joan Claudi Socoro, Francesc Alias, Oriol Guasch
APPLIED SCIENCES-BASEL
(2019)
Article
Computer Science, Information Systems
Xavier Sevillano, Joan Claudi Socoro, Francesc Alias
INFORMATION SCIENCES
(2020)
Article
Chemistry, Analytical
Francesc Alias, Ferran Orga, Rosa Ma Alsina-Pages, Joan Claudi Socoro
Article
Acoustics
Marc Freixes, Francesc Alias, Joan Claudi Socoro
EURASIP JOURNAL ON AUDIO SPEECH AND MUSIC PROCESSING
(2019)
Editorial Material
Chemistry, Multidisciplinary
Francesc Alias, Antonio Bonafonte, Antonio Teixeira
APPLIED SCIENCES-BASEL
(2020)
Article
Chemistry, Analytical
Francesc Alias, Joan Claudi Socoro, Rosa Ma Alsina-Pages
Article
Chemistry, Multidisciplinary
Joan Claudi Socoro, Francesc Alias, Rosa Ma Alsina-Pages
Summary: This paper introduces a clustering method to analyze and categorize the spectro-temporal features of road traffic noise (RTN) collected. The results of the experiments show that the clustering solutions of RTN vary in different environments.
APPLIED SCIENCES-BASEL
(2022)
Article
Chemistry, Multidisciplinary
Marc Freixes, Joan Claudi Socoro, Francesc Alias
Summary: This study analyzes the contribution of vocal tract and glottal source spectral cues in speech generation. The results show that vocal tract cues significantly contribute to the expression of happy and aggressive emotions for [a] vowels, while glottal source spectral cues significantly contribute to [u] vowels.
APPLIED SCIENCES-BASEL
(2022)
Article
Acoustics
Francesc Alias, Rosa Ma. Alsina-Pages
Summary: This study analyzes the impact of COVID-19 lockdown on the acoustic environment of Milan and Rome. In Rome, there is a significant increase in anomalous noise events (ANE) during the lockdown, especially at night and on weekends, despite a decrease in prominent events. In contrast, ANEs decrease during the lockdown in Milan, mostly during the daytime. The intermittency ratio (IR), representing the impact of noise on the population, significantly decreases in most sensing locations during the lockdown, indicating a reduction in the negative impact of noise.
JOURNAL OF THE ACOUSTICAL SOCIETY OF AMERICA
(2022)
Article
Chemistry, Multidisciplinary
Marc Freixes, Luis Joglar-Ongay, Joan Claudi Socoro, Francesc Alias-Pujol
Summary: Currently, articulatory-based three-dimensional source-filter models have limited expressiveness in producing vowels and diphtongs. Glottal inverse filtering (GIF) techniques can be used to identify specific characteristics of the glottal source signal and vocal tract transfer function, allowing for expressive speech synthesis. In this study, a two-phase analysis methodology is introduced for comparing GIF techniques based on a reference dataset. State-of-the-art GIF techniques based on iterative adaptive inverse filtering (IAIF) and quasi closed phase (QCP) approaches are evaluated on the OPENGLOT database, and the results show that QCP-based techniques outperform IAIF-based methods in most error metrics and scenarios.
APPLIED SCIENCES-BASEL
(2023)
Article
Acoustics
Rosa Maria Alsina Pages, Francesc Alias, Patrizia Bellucci, Pier Paolo Cartolano, Ilaria Coppa, Laura Peruzzi, Alessandro Bisceglie, Giovanni Zambon
Proceedings Paper
Computer Science, Theory & Methods
Unai Hernandez-Jayo, Rosa Ma Alsina-Pages, Ignacio Angulo, Francesc Alias
ONLINE ENGINEERING & INTERNET OF THINGS
(2018)