Article
Computer Science, Artificial Intelligence
Carlos Puerto-Santana, Pedro Larranaga, Concha Bielza
Summary: This article introduces asymmetric hidden Markov models with feature saliencies, which are capable of simultaneously determining relevant variables/features and probabilistic relationships between variables during their learning phase. Comparing with other approaches, the proposed models have better or equal fitness and provide further data insights.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Carlos Puerto-Santana, Pedro Larranaga, Concha Bielza
Summary: In a real-life process evolving over time, the relationship between relevant variables may change. Asymmetric hidden Markov models provide a dynamic framework where different inference models can be used for each state of the process. This paper modifies recent asymmetric hidden Markov models to incorporate an asymmetric autoregressive component for continuous variables, allowing the model to choose the optimal order of autoregression. The paper also demonstrates the adaptation of inference, hidden states decoding, and parameter learning for the proposed model. Experimental results with synthetic and real data showcase the capabilities of this new model.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Mathematics, Interdisciplinary Applications
Abdessatar Souissi, El Gheteb Soueidi
Summary: This paper aims to expand on previous research on quantum hidden Markov processes by introducing the concept of entangled hidden Markov processes. These are hidden Markov processes in which the hidden processes themselves are entangled Markov processes. The paper provides an explicit expression for the joint expectation of these processes and demonstrates that the approach also applies to the classical case.
CHAOS SOLITONS & FRACTALS
(2023)
Article
Computer Science, Artificial Intelligence
Samr Ali, Nizar Bouguila
Summary: The hidden Markov model is a key generative machine learning approach in time series data analysis, with recent research focusing on parameter inference and feature selection, applied to tasks such as dynamic texture classification and infrared action recognition.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Physics, Multidisciplinary
Ozgur Danisman, Umay Uzunoglu Kocer
Summary: Hidden Markov models are commonly used for modeling probabilistic structures with latent variables. They assume that observation symbols are conditionally independent and identically distributed, but in practice, this assumption may not always hold. The proposed model introduces a first-order Markov dependency between the current pair of hidden state-emitted observation symbol and the previous pair, which can better capture possible dependencies in real-life scenarios.
PHYSICA A-STATISTICAL MECHANICS AND ITS APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Wentao Fan, Ru Wang, Nizar Bouguila
Summary: This article introduces a method for modeling positive sequential vectors using continuous hidden Markov models. The method uses a mixture of inverted Dirichlet distributions as the emission density and incorporates an unsupervised localized feature selection method, allowing for both positive sequential data modeling and feature selection simultaneously.
PATTERN RECOGNITION
(2021)
Article
Computer Science, Artificial Intelligence
Elizabeth Fons, Paula Dawson, Jeffrey Yau, Xiao-jun Zeng, John Keane
Summary: The financial crisis of 2008 triggered interest in more transparent, rules-based portfolio construction strategies, with smart beta strategies becoming a trend among institutional investors. Researchers have utilized Hidden Markov Models (HMMs) to build a dynamic asset allocation system to manage short-term risk and proposed a smart beta allocation system based on the FSHMM algorithm, showing significant improvement in risk-adjusted returns.
EXPERT SYSTEMS WITH APPLICATIONS
(2021)
Article
Ecology
N. J. Klappstein, L. Thomas, T. Michelot
Summary: The HMM-SSF method provides a versatile framework for analyzing behavior-specific habitat selection, allowing for simultaneous estimation of behavior transitions and habitat selection. It offers a more efficient and general approach compared to previous methods, and can be applied to a wide range of species and systems.
Article
Biochemical Research Methods
Srilakshmi Pattabiraman, Tandy Warnow
Summary: Profile Hidden Markov Models (HMMs) are graphical models that generate finite length sequences from a distribution, widely used in bioinformatics. The construction of profile HMMs is a statistical estimation problem, and it is unknown whether they are statistically identifiable.
IEEE-ACM TRANSACTIONS ON COMPUTATIONAL BIOLOGY AND BIOINFORMATICS
(2021)
Article
Automation & Control Systems
Rahul Singh, Qinsheng Zhang, Yongxin Chen
Summary: In this paper, an algorithm is proposed for estimating the parameters of a time-homogeneous hidden Markov model (HMM) from aggregate observations. The algorithm is built upon the expectation-maximization algorithm and the aggregate inference algorithm, and it exhibits convergence guarantees for both discrete and continuous observations. When the population size is 1, the algorithm is equivalent to the standard Baum-Welch learning algorithm.
Article
Automation & Control Systems
Ali H. Al-gumaei, Muhammad Azam, Manar Amayri, Nizar Bouguila
Summary: Machine learning, a branch of artificial intelligence, focuses on analyzing and interpreting patterns and structures in data to enable autonomous learning and decision-making. Hidden Markov models (HMMs) have recently experienced a resurgence in machine learning and are considered powerful probabilistic models. This paper integrates independent component analysis (ICA) and a bounded multivariate generalized Gaussian mixture model (ICA-BMGGMM) into the HMM approach, overcoming the limitation of assuming independent sources. The proposed models are validated with various applications and outperform other models in terms of performance metrics.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Isidoros Perikos, Spyridon Kardakis, Ioannis Hatzilygeroudis
Summary: The article introduces a novel, interpretable HMM-based method for recognizing sentiments in text, which has been tested under various architectures, training methods, orders, and ensembles, showing competitive performance and outperforming traditional HMM methods.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Management
Maria Luz Gamiz, Nikolaos Limnios, Maria Del Carmen Segovia-Garcia
Summary: This paper discusses the use of hidden Markov models in reliability engineering, where the state of the system is not directly observable. The paper investigates the maximum-likelihood estimation of system reliability and the effectiveness of preventive maintenance strategies. Extensive simulation studies are conducted to evaluate the finite sample performance of the methodology.
EUROPEAN JOURNAL OF OPERATIONAL RESEARCH
(2023)
Article
Engineering, Electrical & Electronic
Laura Piho, Maarja Kruusmaa
Summary: Mapping subsurface flows is challenging due to their inaccessibility and complexity. This paper presents a method using infinite hidden Markov models and inertial measurement unit data to detect features and reconstruct subsurface flow paths. The method is validated on controlled examples and real-world datasets.
IEEE SENSORS JOURNAL
(2022)
Article
Acoustics
Lucas Ondel, Bolaji Yusuf, Lukas Burget, Murat Saraclar
Summary: This study investigates subspace non-parametric models for learning a set of acoustic units from unlabeled speech recordings. By constraining the base-measure of a Dirichlet-Process mixture with a phonetic subspace estimated from other source languages, the learned acoustic units are forced to resemble phones of known source languages. Two models, the Subspace HMM (SHMM) and the Hierarchical-Subspace HMM (H-SHMM), are proposed and applied to three languages. The experimental results show that both subspace models outperform other systems in terms of clustering quality and segmentation accuracy.
IEEE-ACM TRANSACTIONS ON AUDIO SPEECH AND LANGUAGE PROCESSING
(2022)
Review
Engineering, Electrical & Electronic
Zhen-Hua Ling, Shi-Yin Kang, Heiga Zen, Andrew Senior, Mike Schuster, Xiao-Jun Qian, Helen Meng, Li Deng
IEEE SIGNAL PROCESSING MAGAZINE
(2015)
Article
Acoustics
Heiga Zen, Norbert Braunschweiler, Sabine Buchholz, Mark J. F. Gales, Kate Knill, Sacha Krstulovic, Javier Latorre
IEEE TRANSACTIONS ON AUDIO SPEECH AND LANGUAGE PROCESSING
(2012)
Article
Acoustics
Matt Shannon, Heiga Zen, William Byrne
IEEE TRANSACTIONS ON AUDIO SPEECH AND LANGUAGE PROCESSING
(2013)
Article
Engineering, Electrical & Electronic
Keiichi Tokuda, Yoshihiko Nankaku, Tomoki Toda, Heiga Zen, Junichi Yamagishi, Keiichiro Oura
PROCEEDINGS OF THE IEEE
(2013)
Article
Engineering, Electrical & Electronic
Reinhold Haeb-Umbach, Shinji Watanabe, Tomohiro Nakatani, Michiel Bacchiani, Bjoern Hoffmeister, Michael L. Seltzer, Heiga Zen, Mehrez Souden
IEEE SIGNAL PROCESSING MAGAZINE
(2019)
Proceedings Paper
Audiology & Speech-Language Pathology
Ye Jia, Heiga Zen, Jonathan Shen, Yu Zhang, Yonghui Wu
Summary: PnG BERT is a new encoder model for neural TTS that incorporates phoneme and grapheme representations as input, resulting in more natural prosody and accurate pronunciation. Experimental results demonstrate that a neural TTS model pre-trained with PnG BERT outperforms baseline models.
Proceedings Paper
Audiology & Speech-Language Pathology
Nanxin Chen, Yu Zhang, Heiga Zen, Ron J. Weiss, Mohammad Norouzi, Najim Dehak, William Chan
Summary: WaveGrad 2 is a non-autoregressive generative model for text-to-speech synthesis that generates high fidelity audio through an iterative refinement process and allows for a trade-off between inference speed and sample quality by adjusting the number of refinement steps. Experiments show that it approaches the performance of state-of-the-art neural TTS systems.
Proceedings Paper
Audiology & Speech-Language Pathology
Zhehuai Chen, Andrew Rosenberg, Yu Zhang, Heiga Zen, Mohammadreza Ghodsi, Yinghui Huang, Jesse Emond, Gary Wang, Bhuvana Ramabhadran, Pedro J. Moreno
Summary: Semi and self-supervised training techniques can improve speech recognition performance without additional transcribed speech data. This study demonstrates the efficacy of two approaches by leveraging unspoken text and untranscribed audio, reducing word error rate in Indic language voice search tasks by up to 14.4%.
Proceedings Paper
Audiology & Speech-Language Pathology
Isaac Elias, Heiga Zen, Jonathan Shen, Yu Zhang, Ye Jia, R. J. Skerry-Ryan, Yonghui Wu
Summary: This paper introduces Parallel Tacotron 2, a non-autoregressive neural text-to-speech model with a fully differentiable duration model that can learn token-frame alignments and durations automatically. Experimental results show that Parallel Tacotron 2 outperforms baselines in subjective naturalness in several diverse multi-speaker evaluations.
Proceedings Paper
Acoustics
Guangzhi Sun, Yu Zhang, Ron J. Weiss, Yuan Cao, Heiga Zen, Yonghui Wu
2020 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING
(2020)
Proceedings Paper
Acoustics
Guangzhi Sun, Yu Zhang, Ron J. Weiss, Yuan Cao, Heiga Zen, Andrew Rosenberg, Bhuvana Ramabhadran, Yonghui Wu
2020 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING
(2020)
Proceedings Paper
Computer Science, Artificial Intelligence
Antoine Bruguier, Heiga Zen, Arkady Arkhangorodsky
19TH ANNUAL CONFERENCE OF THE INTERNATIONAL SPEECH COMMUNICATION ASSOCIATION (INTERSPEECH 2018), VOLS 1-6: SPEECH RESEARCH FOR EMERGING MARKETS IN MULTILINGUAL SOCIETIES
(2018)
Proceedings Paper
Acoustics
Heiga Zen, Yannis Agiomyrgiannakis, Niels Egberts, Fergus Henderson, Przemyslaw Szczepaniak
17TH ANNUAL CONFERENCE OF THE INTERNATIONAL SPEECH COMMUNICATION ASSOCIATION (INTERSPEECH 2016), VOLS 1-5: UNDERSTANDING SPEECH PROCESSING IN HUMANS AND MACHINES
(2016)
Proceedings Paper
Acoustics
Bo Li, Heiga Zen
17TH ANNUAL CONFERENCE OF THE INTERNATIONAL SPEECH COMMUNICATION ASSOCIATION (INTERSPEECH 2016), VOLS 1-5: UNDERSTANDING SPEECH PROCESSING IN HUMANS AND MACHINES
(2016)
Proceedings Paper
Acoustics
Keiichi Tokuda, Heiga Zen
2016 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING PROCEEDINGS
(2016)
Article
Acoustics
Fazal E. Wahab, Zhongfu Ye, Nasir Saleem, Rizwan Ullah
Summary: This study presents a compact neural model designed in a complex frequency domain for real-time speech enhancement. The proposed model outperforms benchmark models and improves speech quality and intelligibility. The incorporation of attention-gate-based skip connections further enhances the performance.
SPEECH COMMUNICATION
(2024)
