Utterance Verification and Word Spotting 2
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
Context Dependent Anti Subword Modeling for Utterance Verification
Authors:
Padma Ramesh, Lucent Technologies (USA)
Chin-Hui Lee, Lucent Technologies (USA)
Biing-Hwang Juang, Lucent Technologies (USA)
Page (NA) Paper number 880
Abstract:
Utterance verification is used in spoken language dialog systems to reject the speech that does not belong to the task and to correctly recognize the sentences that do. Current verification systems use context dependent (CD) or context independent (CI) subword models and CI anti-subword models. We propose many methods of modeling the CD anti-subword models. We have compared these anti-models and show that the anti-models with the same context have the most separation between the speech that contains the subword and the speech that does not contain the subword. We have also conducted recognition/verification experiments with a two pass verifier and two one pass verification systems to compare the different types of anti-subword models. Our results show that the same context anti-subword models have the best recognition/verification performance.
SL980880.PDF (From Author)
SL980880.PDF (Rasterized)
TOP
Combination of Confidence Measures in Isolated Word Recognition
Authors:
J.G.A. Dolfing, Philips Research Laboratories (Germany)
Andreas Wendemuth, Philips Research Laboratories (Germany)
Page (NA) Paper number 481
Abstract:
In the context of command-and-control applications, we exploit confidence measures in order to classify utterances into two categories: utterances within the vocabulary which are recognized correctly, and other (out-of-vocabulary= OOV and misrecognized) utterances. We investigate the classification error rate (CER) of several classes of confidence measures and transformations based on a database containing 3345 utterances by 50 male and female individuals, employing data-independent and data-dependent measures. The transformations we investigated include mapping to single confidence measures, LDA-transformed measures, and other linear combinations of these measures. These combinations are computed by means of neural networks trained with Bayes-optimal, and with Gardner-Derrida-optimal criteria. Compared to a recognition system without confidence measures, the selection of (various combinations of) confidence measures, and the selection of suitable neural network architectures and training methods, continuously improves the CER from 16.7% to 6.6% (-60% relative). Furthermore, a linear perceptron generalizes better than a non-linear backpropagation network.
SL980481.PDF (From Author)
SL980481.PDF (Rasterized)
TOP
Confidence Measures for HMM-based Speech Recognition
Authors:
Daniel Willett, Gerhard-Mercator-University, Duisburg (Germany)
Andreas Worm, Gerhard-Mercator-University, Duisburg (Germany)
Christoph Neukirchen, Gerhard-Mercator-University, Duisburg (Germany)
Gerhard Rigoll, Gerhard-Mercator-University, Duisburg (Germany)
Page (NA) Paper number 525
Abstract:
In this paper, we describe our work on the field of confidence measures for HMM-based speech recognition. Confidence measures are a means of estimating the recognition reliability for single words of the recognizer output. The possible applications of such measures are manifold. We present our experiments with well known approaches and propose some new ones. Particularly, we propose to combine the mere acoustical measures with language model-based ones for continuous speech recognition that involves a stochastic language model. This slightly improves the acoustical measures and preserves their advantage of being computationally very cheap. Experiments are carried out on a German isolated word recognition system and on continuous speech recognition systems for the Resource Management database and the Wall Street Journal WSJ0 task.
SL980525.PDF (From Author)
SL980525.PDF (Rasterized)
TOP
Vocabulary-Independent Word Confidence Measure Using Subword Features
Authors:
Li Jiang, Microsoft Research (USA)
Xuedong Huang, Microsoft Research (USA)
Page (NA) Paper number 625
Abstract:
This paper discusses how to compute word-level confidence measures based on sub-word features for large-vocabulary speaker-independent speech recognition. The performance of confidence measure using features at word, phone and senone level is experimentally studied. A framework of transformation function based system using sub-word features is proposed for high performance confidence estimation. In this system, discriminative training is used to optimize the parameters of the transformation function. In comparison to the baseline, experiments show that the proposed system reduces the equal error rate by 15%, with up to 40% false acceptance error reduction at various fixed false rejection rate. The combination of multiple features under the proposed framework is also discussed.
SL980625.PDF (From Author)
SL980625.PDF (Rasterized)
TOP
A New Confidence Measure Based on Rank-Ordering Subphone Scores
Authors:
Qiguang Lin, IBM T. J. Watson Research Center (USA)
Subrata Das, IBM T. J. Watson Research Center (USA)
David Lubensky, IBM T. J. Watson Research Center (USA)
Michael Picheny, IBM T. J. Watson Research Center (USA)
Page (NA) Paper number 806
Abstract:
This paper presents a new approach to measuring how confidently a word has been correctly recognized, or confidence measure. The approach consists of three major steps: (1) standard decoding; (2) forced Viterbi alignment (needed for stack decoders); and (3) rank-ordering of the subphone scores. More specifically, from the aligned sentence the third step computes the likelihood scores of the hypothesized subphone and all other competing subphones. A list of the subphones is generated in the descending order of the scores and a rank is assigned to the hypothesized subphone according to its positioning. Additional processing of selective weighting and upper-bound limiting is applied to minimize contamination of rank computations by bad segments or by highly-variable phones. The obtained rank is then used as the confidence measure. Results of word rejection experiments show that the new approach outperforms other measures such as whole-word scores by reducing the equal error rate from 32% to 20%.
SL980806.PDF (From Author)
SL980806.PDF (Rasterized)
TOP
Speaking-Style Dependent Lexicalized Filler Model for Key-Phrase Detection and Verification
Authors:
Tatsuya Kawahara, Kyoto University (Japan)
Kentaro Ishizuka, Kyoto University (Japan)
Shuji Doshita, Kyoto University (Japan)
Chin-Hui Lee, Bell Laboratories (USA)
Page (NA) Paper number 761
Abstract:
A task-independent filler modeling for robust key-phrase detection and verification is proposed. Instead of assuming task-specific lexical knowledge, our model is designed to characterize phrases depending on the speaking-style, thus can be trained with large corpora of different but similar tasks. We present two implementations of the portable and general model. The dialogue-style dependent model trained with the ATIS corpus is used as a filler and shown to be effective in detection-based speech understanding on different dialogue applications. The lecture-style dependent filler model trained with transcriptions of various oral presentations also improves the verification of key-phrases uttered during lectures.