Speaker Adaptation 3
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
On-line Hierarchical Transformation of Hidden Markov Models for Speaker Adaptation
Authors:
Jen-Tzung Chien, National Cheng Kung University (Taiwan)
Page (NA) Paper number 102
Abstract:
This paper presents a novel framework of on-line hierarchical transformation of hidden Markov models (HMM's) for speaker adaptation. Our aim is to incrementally transform (or adapt) all the HMM parameters to a new speaker even though part of HMM units are unseen in adaptation data. The transformation paradigm is formulated according to the approximate Bayesian estimate, which the prior statistics and the transformation parameters are incrementally updated for each consecutive adaptation data. Using this formulation, the updated prior statistics and the current block of data are sufficient for on-line transformation. Further, we establish a hierarchical tree of HMM's and use it to dynamically control the transformation sharing for each HMM unit. In the speaker adaptation experiments, we demonstrate the superiority of proposed on-line transformation to other method.
SL980102.PDF (From Author)
SL980102.PDF (Rasterized)
TOP
High-Speed Speaker Adaptation Using Phoneme Dependent Tree-Structured Speaker Clustering
Authors:
Motoyuki Suzuki, Computer Center / Graduate school of Information Sciences, Tohoku Univ. (Japan)
Toshiaki Abe, Graduate school of Engineering, Tohoku Univ. (Japan)
Hiroki Mori, Graduate school of Engineering, Tohoku Univ. (Japan)
Shozo Makino, Computer Center / Graduate school of Information Sciences, Tohoku Univ. (Japan)
Hirotomo Aso, Graduate school of Engineering, Tohoku Univ. (Japan)
Page (NA) Paper number 992
Abstract:
The tree-structured speaker clustering was proposed as a high-speed speaker adaptation method. It can select the model which is most similar to a target speaker. However, this method does not consider speaker difference dependent on phoneme class. In this paper, we propose a speaker adaptation method based on speaker clustering by taking speaker difference dependent on phoneme class into account. The experimental results showed that the new method gave a better performance than the original method. Furthermore, we propose the improved method which use a tree-structure of a similar phoneme as the substitute for the phoneme which does not appear in the adaptation data. From the experimental results, the improved method gave a better performance than the method previously proposed.
SL980992.PDF (From Author)
SL980992.PDF (Rasterized)
TOP
The Use of Confidence Measures in Unsupervised Adaptation of Speech Recognizers
Authors:
Tasos Anastasakos, Motorola, Lexicus Division (USA)
Sreeram V. Balakrishnan, Motorola, Lexicus Division (USA)
Page (NA) Paper number 599
Abstract:
Confidence estimation of the output hypothesis of a speech recognizer offers a way to assess the probability that the recognized words are correct. This work investigates the application of confidence scores for selection of speech segments in unsupervised speaker adaptation. Our approach is motivated by initial experiments that show that the use of mis-labeled data has a significant cost in the performance of particular adaptation schemes. We focus on a rapid self-adaptation scenario that uses only a few seconds of adaptation data. The adaptation algorithm is based on an extension to the MLLR transformation method that can be applied to the observation vectors. We present experimental results of this work on the ARPA WSJ large vocabulary dictation task.
SL980599.PDF (From Author)
SL980599.PDF (Rasterized)
TOP
Speaker Normalization with All-Pass Transforms
Authors:
John McDonough, Center for Language and Speech Processing, The Johns Hopkins University (USA)
William Byrne, Center for Language and Speech Processing, The Johns Hopkins University (USA)
Xiaoqiang Luo, Center for Language and Speech Processing, The Johns Hopkins University (USA)
Page (NA) Paper number 869
Abstract:
Speaker normalization is a process in which the short-time features of speech from a given speaker are transformed so as to better match some speaker independent model. Vocal tract length normalization (VTLN) is a popular speaker normalization scheme wherein the frequency axis of the short-time spectrum associated with a speaker's speech is rescaled or warped prior to the extraction of cepstral features. In this work, we develop a novel speaker normalization scheme by exploiting the fact that frequency domain transformations similar to that inherent in VTLN can be accomplished entirely in the cepstral domain through the use of conformal maps. We propose a class of such maps, designated all-pass transforms for reasons given hereafter, and in a set of speech recognition experiments conducted on the Switchboard Corpus demonstrate their capacity to achieve word error rate reductions of 3.7% absolute.
SL980869.PDF (From Author)
SL980869.PDF (Rasterized)
TOP
Toward On-Line Learning of Chinese Continuous Speech Recognition System
Authors:
Rong Zheng, Speech Recognition Lab, Dept. of Electrical Engr. , Tsinghua University (China)
Zuoying Wang, Speech Recognition Lab, Dept. of Electrical Engr. , Tsinghua University (China)
Page (NA) Paper number 276
Abstract:
In this paper, we presented an integrated on-line learning scheme, which combined the state-of-art speaker normalization and adaptation techniques to improve the performance of our large vocabulary Chinese continuous speech recognition (CSR) system. We used VTLN to remove inter-speaker variation in both training and testing stage. To facilitate dynamic transformation scale determination, we devised a tree-based transformation method as the key component of our incremental adaptation. Experiments shows that the combined scheme of on-line learning (incremental & unsupervised) system, which gives approximately 22~26% error reduction rate, was proved to be better than either method when used separately at 18.34% and 2.7%.
SL980276.PDF (From Author)
SL980276.PDF (Rasterized)
TOP
The CHAM Model of Hyperarticulate Adaptation During Human-Computer Error Resolution
Authors:
Sharon L. Oviatt, Oregon Graduate Institute (USA)
Page (NA) Paper number 49
Abstract:
When using interactive systems, people adapt their speech during attempts to resolve system recognition errors. This paper summarizes the two-stage Computer-elicited Hyperarticulate Adaptation Model (CHAM), which accounts for systematic changes in human speech during interactive error handling. According to CHAM, Stage I adaptation is manifest as a singular change involving the increased duration of speech and pauses. This change is associated with a moderate degree of hyperarticulation, which occurs during a low rate of system errors. In contrast, State II adaptations are associated with more extreme hyperarticulation during a high system error rate. It entails change in multiple features of speech - including duration, articulation, intonation pattern, fundamental frequency and amplitude. This paper summarizes the empirical findings and linguistic theory upon which CHAM is based, as well as the model's main predictions. Finally, the implications of CHAM are discussed for designing future interactive systems with improved error handling.