Speaker and Language Recognition 4
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 The Convergence Of Gaussian Mixture Models: Improvements Through Vector Quantization
Authors:
James Moody, Queensland University of Technology (Australia)
Stefan Slomka, Queensland University of Technology (Australia)
Jason Pelecanos, Queensland University of Technology (Australia)
Sridha Sridharan, Queensland University of Technology (Australia)
Page (NA) Paper number 667
Abstract:
This paper studies the reliance of a Gaussian Mixture Model (GMM) based closed-set Speaker Identification system on model convergence and describes methods to improve this convergence. It shows that the reason why the Vector Quantisation GMMs (VQGMMs) outperform a simple GMM is mainly due to decreasing the complexity of the data during training. In addition, it is shown that the VQGMM system is less computationally complex than the traditional GMM, yielding a system which is quicker to train and which gives higher performance. We also investigate four different VQ distance measures which can be used in the training of a VQGMM and compare their respective performances. It is found that the improvements gained by the VQGMM is only marginally dependant on the distance measure.
SL980667.PDF (From Author)
SL980667.PDF (Rasterized)
TOP
Modeling Dynamic Prosodic Variation for Speaker Verification
Authors:
Kemal Sönmez, SRI International (USA)
Elizabeth Shriberg, SRI International (USA)
Larry Heck, Nuance Communications (USA)
Mitchel Weintraub, SRI International (USA)
Page (NA) Paper number 920
Abstract:
Statistics of frame-level pitch have recently been used in speaker recognition systems with good results. Although they convey useful long-term information about a speaker's distribution of f0 values, such statistics fail to capture information about local dynamics in intonation that characterize an individual's speaking style. In this work, we take a first step toward capturing such suprasegmental patterns for automatic speaker verification. Specifically, we model the speaker's f0 movements by fitting a piecewise linear model to the f0 track to obtain a stylized f0 contour. Parameters of the model are then used as statistical features for speaker verification. We report results on 1998 NIST speaker verification evaluation. Prosody modeling improves the verification performance of a cepstrum-based Gaussian mixture model system (as measured by a task-specific Bayes risk) by 10%.
SL980920.PDF (From Author)
SL980920.PDF (Rasterized)
TOP
Blind Clustering of Speech Utterances Based on Speaker and Language Characteristics
Authors:
Douglas A. Reynolds, MIT Lincoln Laboratory (USA)
Elliot Singer, MIT Lincoln Laboratory (USA)
Beth A. Carlson, MIT Lincoln Laboratory (USA)
Gerald C. O'Leary, MIT Lincoln Laboratory (USA)
Jack J. McLaughlin, MIT Lincoln Laboratory (USA)
Marc A. Zissman, MIT Lincoln Laboratory (USA)
Page (NA) Paper number 610
Abstract:
Classical speaker and language recognition techniques can be applied to the classification of unknown utterances by computing the likelihoods of the utterances given a set of well trained target models. This paper addresses the problem of grouping unknown utterances when no information is available regarding the speaker or language classes or even the total number of classes. Approaches to blind message clustering are presented based on conventional hierarchical clustering techniques and an integrated cluster generation and selection method called the d* algorithm. Results are presented using message sets derived from the Switchboard and Callfriend corpora. Potential applications include automatic indexing of recorded speech corpora by speaker/language tags and automatic or semiautomatic selection of speaker specific speech utterances for speaker recognition adaptation.
SL980610.PDF (From Author)
SL980610.PDF (Rasterized)
TOP
Spoken Language Identification Using The SpeechDat Corpus
Authors:
Diamantino Caseiro, INESC/IST (Portugal)
Isabel M. Trancoso, INESC/IST (Portugal)
Page (NA) Paper number 1093
Abstract:
Current language identification systems vary significantly in their complexity. The systems that use higher level linguistic information have the best performance. Nevertheless, that information is hard to collect for each new language. The system presented in this paper is easily extendable to new languages because it uses very little linguistic information. In fact, the presented system needs only one language specific phone recogniser (in our case the Portuguese one), and is trained with speech from each of the other languages. With the SpeechDat-M corpus, with 6 European languages (English, French, German, Italian, Portuguese and Spanish) our system achieved an identification rate of 83.4% on 5-second utterances, this result shows an improvement of 5% over our previous version, mainly through the use of a neural network classifier. Both the baseline and the full system were implemented in realtime.
SL981093.PDF (From Author)
SL981093.PDF (Rasterized)
TOP
Automatic Language Identification with Perceptually Guided Training and Recurrent Neural Networks
Authors:
Jerome Braun, Computer Science Department, University of Massachusetts Lowell (USA)
Haim Levkowitz, Computer Science Department, University of Massachusetts Lowell (USA)
Page (NA) Paper number 405
Abstract:
We present a novel approach to Automatic Language Identification (LID). We propose Perceptually Guided Training (PGT), a novel LID training method, involving identification of utterance parts which are particularly significant perceptually for the language identification process, and exploitation of these Perceptually Significant Regions (PSRs) to guide the LID training process. Our approach involves a Recurrent Neural Network (RNN) as the main mechanism. We propose that, because of the long-range intra-utterance acoustical context significance in LID, RNNs are particularly suitable for the LID task. Our approach does not require phonetic labeling or transcription of the training corpus. LIREN/PGT, the LID system we developed, incorporates our approach. Our LID experiments were on English, German, and Mandarin Chinese, using the OGI-TS corpus.
SL980405.PDF (From Author)
SL980405.PDF (Scanned)
TOP
On the Importance of Components of the Modulation Spectrum for Speaker Verification
Authors:
Sarel van Vuuren, Oregon Graduate Institute of Science and Technology (USA)
Hynek Hermansky, Oregon Graduate Institute of Science and Technology (USA)
Page (NA) Paper number 631
Abstract:
We provide an analysis of the relative importance of components of the modulation spectrum for speaker verification. The aim is to remove less relevant components and reduce system sensitivity to acoustic disturbances while improving verification accuracy. Spectral components between about 0.1Hz and 10Hz are found to contain the most useful speaker information. We discuss this result in the context of RASTA processing and cepstral mean subtraction. When compared to cepstral mean subtraction that retains components up to 50Hz, lowpass filtering to 10Hz with downsampling by 75 percent is found to significantly improve robustness in mismatched conditions. The downsampling results in a large computational savings.