Neural Networks, Fuzzy and Evolutionary Methods 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
Fuzzy-Integration Based Normalization for Speaker Verification
Authors:
Tuan Pham, Faculty of Information Sciences & Engineering, University of Canberra (Australia)
Michael Wagner, Faculty of Information Sciences & Engineering, University of Canberra (Australia)
Page (NA) Paper number 953
Abstract:
Similarity or likelihood normalization techniques are important for speaker verification systems as they help to alleviate the variations in the speech signals. In the conventional normalization, the a priori probabilities of the cohort speakers are assumed to be equal. From this standpoint, we apply the theory of fuzzy measure and fuzzy integral to combine the likelihood values of the cohort speakers in which the assumption of equal a priori probabilities is relaxed. This approach replaces the conventional normalization term by the fuzzy integral which acts as a non-linear fusion of the similarity measures of an utterance assigned to the cohort speakers. We illustrate the performance of the proposed approach by testing the speaker verification system with both the conventional and the fuzzy algorithms using the commercial speech corpus TI46. The results in terms of the equal error rates show that the speaker verification system using the fuzzy integral is more flexible and more favorable than the conventional normalization method.
SL980953.PDF (From Author)
SL980953.PDF (Rasterized)
TOP
Improving The Generalization Performance Of The MCE/GPD Learning
Authors:
Hiroshi Shimodaira, Japan Advanced Institute of Science and Technology (Japan)
Jun Rokui, Japan Advanced Institute of Science and Technology (Japan)
Mitsuru Nakai, Japan Advanced Institute of Science and Technology (Japan)
Page (NA) Paper number 795
Abstract:
A novel method to prevent the over-fitting effect and improve the generalization performance of the Minimum Classification Error (MCE) / Generalized Probabilistic Descent (GPD) learning is proposed. The MCE/GPD method, which is one of the newest discriminative-learning approaches proposed by Katagiri and Juang in 1992, results in better recognition performance in various areas of pattern recognition than the maximum-likelihood (ML) based approach where a posteriori probabilities are estimated. Despite its superiority in recognition performance, it still suffers from the problem of over-fitting to the training samples as it is with other learning algorithms. In the present study, a regularization technique is employed to the MCE method to overcome this problem. Feed-forward neural networks are employed as a recognition platform to evaluate the recognition performance of the proposed method. Recognition experiments are conducted on several sorts of datasets. The proposed method shows better generalization performance than the original one
SL980795.PDF (From Author)
SL980795.PDF (Rasterized)
TOP
Acoustic Speech Recognition Model by Neural Net Equation with Competition and Cooperation
Authors:
Tetsuro Kitazoe, Miyazaki University (Japan)
Tomoyuki Ichiki, Miyazaki University (Japan)
Sung-Ill Kim, Miyazaki University (Japan)
Page (NA) Paper number 965
Abstract:
The equation of neural nets for stereo vision is applied to speech recognition. We use Coupled Pattern Recognition (CPR) equation which has been shown to organize depth perception very well through competition and cooperation. We construct Gaussian probability density function for each phoneme from a number of training data. The input data to be recognized are compared to the pdf's and the similarity measures are obtained for each phoneme. The CPR equation develops neuron activities by receiving the similarity measures as input. A recognition is achieved when the activities arrive at a stable states. The recognition rates for 25 Japanese phoneme are 74.75% in average which is compared to 71.53% Hidden Markov Model. A certain technical improvement is applied to our neuron model, by dividing data of a phoneme into two part, one for the former frames, the other for the latter frames.A remarkable improvement is obtained with average recognition rate of 79.79%.
SL980965.PDF (Scanned)
TOP
Improved Surname Pronunciations Using Decision Trees
Authors:
Julie Ngan, Institute for Signal and Information Processing, Mississippi State University (USA)
Aravind Ganapathiraju, Institute for Signal and Information Processing, Mississippi State University (USA)
Joseph Picone, Institute for Signal and Information Processing, Mississippi State University (USA)
Page (NA) Paper number 384
Abstract:
Proper noun pronunciation generation is a particularly challenging problem in speech recognition since a large percentage of proper nouns often defy typical letter-to-sound conversion rules. In this paper, we present decision tree methods which outperform neural network techniques. Using the decision tree method, we have achieved an overall error rate of 45.5%, which is a 35% reduction over the previous techniques. Our best system is a binary decision tree that uses a context length of 3 and employs information gain ratio as the splitting rule.