Hidden Markov Model Techniques 1
HomeFull List of Titles
1: ICSLP'98 Proceedings
Keynote SpeechesText-To-Speech Synthesis 1Spoken Language Models and Dialog 1Prosody and Emotion 1Hidden Markov Model Techniques 1Speaker and Language Recognition 1Multimodal Spoken Language Processing 1Isolated Word RecognitionRobust Speech Processing in Adverse Environments 1Spoken Language Models and Dialog 2Articulatory Modelling 1Talking to Infants, Pets and LoversRobust Speech Processing in Adverse Environments 2Spoken Language Models and Dialog 3Speech Coding 1Articulatory Modelling 2Prosody and Emotion 2Neural Networks, Fuzzy and Evolutionary Methods 1Utterance Verification and Word Spotting 1 / Speaker Adaptation 1Text-To-Speech Synthesis 2Spoken Language Models and Dialog 4Human Speech Perception 1Robust Speech Processing in Adverse Environments 3Speech and Hearing Disorders 1Prosody and Emotion 3Spoken Language Understanding Systems 1Signal Processing and Speech Analysis 1Spoken Language Generation and Translation 1Spoken Language Models and Dialog 5Segmentation, Labelling and Speech Corpora 1Multimodal Spoken Language Processing 2Prosody and Emotion 4Neural Networks, Fuzzy and Evolutionary Methods 2Large Vocabulary Continuous Speech Recognition 1Speaker and Language Recognition 2Signal Processing and Speech Analysis 2Prosody and Emotion 5Robust Speech Processing in Adverse Environments 4Segmentation, Labelling and Speech Corpora 2Speech Technology Applications and Human-Machine Interface 1Large Vocabulary Continuous Speech Recognition 2Text-To-Speech Synthesis 3Language Acquisition 1Acoustic Phonetics 1Speaker Adaptation 2Speech Coding 2Hidden Markov Model Techniques 2Multilingual Perception and Recognition 1Large Vocabulary Continuous Speech Recognition 3Articulatory Modelling 3Language Acquisition 2Speaker and Language Recognition 3Text-To-Speech Synthesis 4Spoken Language Understanding Systems 4Human Speech Perception 2Large Vocabulary Continuous Speech Recognition 4Spoken Language Understanding Systems 2Signal Processing and Speech Analysis 3Human Speech Perception 3Speaker Adaptation 3Spoken Language Understanding Systems 3Multimodal Spoken Language Processing 3Acoustic Phonetics 2Large Vocabulary Continuous Speech Recognition 5Speech Coding 3Language Acquisition 3 / Multilingual Perception and Recognition 2Segmentation, Labelling and Speech Corpora 3Text-To-Speech Synthesis 5Spoken Language Generation and Translation 2Human Speech Perception 4Robust Speech Processing in Adverse Environments 5Text-To-Speech Synthesis 6Speech Technology Applications and Human-Machine Interface 2Prosody and Emotion 6Hidden Markov Model Techniques 3Speech and Hearing Disorders 2 / Speech Processing for the Speech and Hearing Impaired 1Human Speech ProductionSegmentation, Labelling and Speech Corpora 4Speaker and Language Recognition 4Speech Technology Applications and Human-Machine Interface 3Utterance Verification and Word Spotting 2Large Vocabulary Continuous Speech Recognition 6Neural Networks, Fuzzy and Evolutionary Methods 3Speech Processing for the Speech-Impaired and Hearing-Impaired 2Prosody and Emotion 72: SST Student Day
SST Student Day - Poster Session 1SST Student Day - Poster Session 2Author Index
A B C D E F G H I
J K L M N O P Q R
S T U V W X Y Z
Multimedia Files
Nonreciprocal Data Sharing in Estimating HMM Parameters
Authors:
Xiaoqiang Luo, CLSP, The Johns Hopkins University (USA)
Frederick Jelinek, CLSP, The Johns Hopkins University (USA)
Page (NA) Paper number 365
Abstract:
Parameter tying is often used in large vocabulary continuous speech recognition (LVCSR) systems to balance the model resolution and generalizability. However, one consequence of tying is that the differences among tied constructs are ignored. Parameter tying can be alternatively viewed as reciprocal data sharing in that a tied construct uses data associated with all others in its tied-class. To capture the fine difference among tied constructs, we propose to use nonreciprocal data sharing (NRDS) when estimating HMM parameters. In particular, when estimating Gaussian parameters for a HMM state, contributions from other acoustically similar HMM states will be weighted, thus allowing different statistics to govern different states. Data sharing weights are optimized using cross-validation. It can be shown that the objective function for cross-validation is a sum of rational functions and can be efficiently optimized by the growth-transform. Our results on Switchboard show that NRDS reduces the word error rate (WER) significantly compared with a state-of-art baseline system using HMM state-tying.
SL980365.PDF (From Author)
SL980365.PDF (Rasterized)
TOP
Data-Driven Extensions to HMM Statistical Dependencies
Authors:
Jeff A. Bilmes, ICSI/U.C. Berkeley (USA)
Page (NA) Paper number 894
Abstract:
In this paper, a new technique is introduced that relaxes the HMM conditional independence assumption in a principled way. Without increasing the number of states, the modeling power of an HMM is increased by including only those additional probabilistic dependencies (to the surrounding observation context) that are believed to be both relevant and discriminative. Conditional mutual information is used to determine both relevance and discriminability. Extended Gaussian-mixture HMMs and new EM update equations are introduced. In an isolated word speech database, results show an average 34% word error improvement over an HMM with the same number of states, and a 15% improvement over an HMM with a comparable number of parameters.
SL980894.PDF (From Author)
SL980894.PDF (Rasterized)
TOP
Use of High-Level Linguistic Constraints for Constructing Feature-Based Phonological Model in Speech Recognition
Authors:
Jiping Sun, Department of Electrical and Computer Engineering, University of Waterloo (Canada)
Li Deng, Department of Electrical and Computer Engineering, University of Waterloo (Canada)
Page (NA) Paper number 43
Abstract:
Modeling phonological units of speech is a critical issue in speech recognition. In this paper, we report our recent development of an overlapping feature-based phonological model which gives long-span contextual dependency. We extend our earlier work by incorporating high-level linguistic constraints in automatic construction of the feature overlapping patterns. The main linguistic information explored includes morpheme, syllable, syllable constituent categories and word stress markers. We describe a consistent computational framework developed for the construction of the feature-based model, and discuss use of the model as the HMM state topology for speech recognizers.