Large Vocabulary Continuous Speech Recognition 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
Grammatical Word Graph Re-Generation for Spontaneous Speech Recognition
Authors:
Hajime Tsukada, ATR Interpreting Telecommunications Research Laboratories (Japan)
Hirofumi Yamamoto, ATR Interpreting Telecommunications Research Laboratories (Japan)
Toshiyuki Takezawa, ATR Interpreting Telecommunications Research Laboratories (Japan)
Yoshinori Sagisaka, ATR Interpreting Telecommunications Research Laboratories (Japan)
Page (NA) Paper number 485
Abstract:
We propose a novel recognition method for generating an accurate grammatical word-graph allowing grammatical deviations. Our method uses both an n-gram and a grammar-based statistical language model and aligns utterances with the grammar by adding deviation information during the search process. Our experiments confirm that the word-graph obtained by our proposed method is superior to the one obtained by only using the n-gram with the same word-graph density. In addition, our recognition method can search enormous hypotheses more efficiently than the conventional word-graph based search method.
SL980485.PDF (From Author)
SL980485.PDF (Rasterized)
TOP
Compression Algorithm Of Trigram Language Models Based On Maximum Likelihood Estimation
Authors:
Norimichi Yodo, Graduate School of Information Science, Nara Institute of Science and Technology (Japan)
Kiyohiro Shikano, Graduate School of Information Science, Nara Institute of Science and Technology (Japan)
Satoshi Nakamura, Graduate School of Information Science, Nara Institute of Science and Technology (Japan)
Page (NA) Paper number 716
Abstract:
In this paper we propose an algorithm for reducing the size of back-off N-gram models, with less affecting its performance than the traditional cutoff method. The algorithm is based on the Maximum Likelihood (ML) estimation and realizes an N-gram language model with a given number of N-gram probability parameters that minimize the training set perplexity. To confirm the effectiveness of our algorithm, we apply it to trigram and bigram models, and the experiments in terms of perplexity and word error rate in a dictation system are carried out.
SL980716.PDF (From Author)
SL980716.PDF (Rasterized)
TOP
Morphological Modeling of Word Classes for Language Models
Authors:
Ulla Uebler, Bavarian Research Center for Knowledge Based Systems (Germany)
Heinrich Niemann, Bavarian Research Center for Knowledge Based Systems (Germany)
Page (NA) Paper number 338
Abstract:
It is well known that good language models improve performance of speech recognition. One requirement for the estimation of language models is a sufficient amount of texts of the application domain. If not all words of the domain occur in the training texts for language models, a way must be found to model these words adequately. In this paper we report on a new approach of building word classes for language modeling in the bilingual (German, Italian) SpeeData project. The main idea is to classify words according to their morphological properties. Therefore we decompose words into their morphological units and put the words with the same prefix or suffix into the same class. Since morphological decomposition is error prone for unknown word stems, we also decomposed words by counting beginnings and endings of different length and used these subunits like prefixes and suffixes. The advantage of this approach is that it can be carried out automatically. We achieved a reduction in error rate from 9.83 % to 5.77 % for morphological decomposition and 5.99 % for automatical decomposition which can be performed without any morphological knowledge.
SL980338.PDF (From Author)
SL980338.PDF (Rasterized)
TOP
A Comparative Study Between Polyclass and Multiclass Language Models
Authors:
Imed Zitouni, LORIA / INRIA-Lorraine (France)
Kamel Smaïli, LORIA / INRIA-Lorraine (France)
Jean-Paul Haton, LORIA / INRIA-Lorraine (France)
Sabine Deligne, ATR-ITL (Japan)
Frédéric Bimbot, IRISA-CNRS/INRIA (France)
Page (NA) Paper number 498
Abstract:
In this paper, we introduce the concept of Multiclass for language modeling and we compare it to the Polyclass model. The originality of the Multiclass is its capability to parse a string of class/tags into variable length independent sequences. A few experimental tests were carried out on a class corpus extracted from the French "Le Monde" word corpus labeled automatically. This corpus contains a set of 43 million of words. In our experiments, Multiclass outperform first-order Polyclass but are slightly outperformed by second-order Polyclass.
SL980498.PDF (From Author)
SL980498.PDF (Rasterized)
TOP
Log-Linear Interpolation Of Language Models
Authors:
Dietrich Klakow, Philips Research Laboratories (Germany)
Page (NA) Paper number 522
Abstract:
Combining different language models is an important task. Linear interpolation is the established method to do this. We will present a new method called log-linear interpolation (LLI) which combines the simplicity of linear interpolation with essential parts from maximum-entropy models by linearly interpolating the scores of different models. The first series of experiments focuses on adaptation. Unigram, bigram and trigram models trained on NAB are combined with unigram and bigram models trained on a small domain specific corpus. LLI compares favorably with linear interpolation. The second series combines bigram and distance-bigram models. Here, relative improvements are larger (~20% in perplexity). This task seems to be the ideal application of LLI. To further scrutinize the method, first frequent pairs of words are joined to phrases and next, bigram and distance-bigram are combined by LLI. This experiment yields perplexities just 2.5% above the original trigram perplexity.
SL980522.PDF (From Author)
SL980522.PDF (Rasterized)
TOP
The Applicability of Adaptive Language Modelling for the Broadcast News Task
Authors:
Philip Clarkson, Cambridge University Engineering Department (U.K.)
Tony Robinson, Cambridge University Engineering Department (U.K.)
Page (NA) Paper number 962
Abstract:
Adaptive language models have consistently been shown to lead to a significant reduction in language model perplexity compared to the equivalent static trigram model on many data sets. When these language models have been applied to speech recognition, however, they have seldom resulted in a corresponding reduction in word error rate. This paper will investigate some of the possible reasons for this apparent discrepancy, and will explore the circumstances under which adaptive language models can be useful. We will concentrate on cache-based and mixture-based models and their use on the Broadcast News task.