Speaker and Language 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
Speaker Detection in Broadcast Speech Databases
Authors:
Aaron E. Rosenberg, AT&T Labs Research (USA)
Ivan Magrin-Chagnolleau, AT&T Labs Research (USA)
S. Parthasarathy, AT&T Labs Research (USA)
Qian Huang, AT&T Labs Research (USA)
Page (NA) Paper number 202
Abstract:
Experiments have been carried out to assess the feasibility of detecting target speaker segments in multi-speaker broadcast databases, The experimental database consists of NBC Nightly News broadcasts. The target speaker is the news anchor, Tom Brokaw. Gaussian mixture models are constructed from labelled training data for the target speaker as well as background models for other speakers,, commercials, and music. Four labelled 30-min. broadcasts are used ffor testing. Mel-frequency cepstral features, augmented by delta cepstral features, are calculated over 20 msec. windows shifted every 10 msecs. through a broadcast. Likelihood ratio scores are calculated for each test frame averaged over blocks of frames with a specified duration. The block scores are input to a detection routine which returns estimates of target segment boundaries. The range of best results obtained over the test broadcasts is 82% to 100% detection of target segments with segment frame accuracy ranging from 86% to 95%. 0 to 2 false alarm segments are detected over each 30 min. broadcast.
SL980202.PDF (From Author)
SL980202.PDF (Rasterized)
TOP
Multilateral Techniques for Speaker Recognition
Authors:
Eluned S. Parris, Ensigma Ltd. (U.K.)
Michael J. Carey, Ensigma Ltd. (U.K.)
Page (NA) Paper number 444
Abstract:
Speaker recognition is usually accomplished by building a set of models from speech of a known speaker, training data, and subsequently using a pattern matching algorithm to score the speech from an unknown speaker, test data. In this paper we discard the notion of train and test data in speaker recognition and introduce the multilateral scoring technique. This technique comprises building speaker models on material for the known speaker and matching the unknown speaker data to these models, the traditional approach to speaker recognition. The resultant scores are fused with an equivalent set of scores produced by matching the known speaker utterance to models built on the unknown speaker data. Significant improvements have been achieved using this technique on the NIST 1996, 1997 and 1998 Speaker Recognition Evaluation data. Results are presented for two speaker recognition systems, the first based on Hidden Markov models and the second based on Gaussian Mixture models.
SL980444.PDF (From Author)
SL980444.PDF (Rasterized)
TOP
Real Time Speaker Indexing Based on Subspace Method - Application to TV News Articles and Debate
Authors:
Masafumi Nishida, Ryukoku University (Japan)
Yasuo Ariki, Ryukoku University (Japan)
Page (NA) Paper number 125
Abstract:
In this paper, we propose a method to extract and verify individual speaker utterance using a subspace method. This method can extract speech section of the same speaker by repeating speaker verification between the present speech section and the immediately previous speech section. The speaker models are automatically trained in the verification process without constructing speaker templates in advance. As a result, this speaker verification method is applied to speaker indexing. In this study, announcer utterances are automatically extracted from news speech data which includes reporter or interviewer utterances. Also extracted automatically are the utterances of each participator in debate program broadcasted on TV.
SL980125.PDF (From Author)
SL980125.PDF (Rasterized)
TOP
SHEEP, GOATS, LAMBS and WOLVES: A Statistical Analysis of Speaker Performance in the NIST 1998 Speaker Recognition Evaluation
Authors:
George Doddington, NIST (USA)
Walter Liggett, NIST (USA)
Alvin Martin, NIST (USA)
Mark Przybocki, NIST (USA)
Douglas A. Reynolds, MIT Lincoln Laboratory (USA)
Page (NA) Paper number 608
Abstract:
Performance variability in speech and speaker recognition systems can be attributed to many factors. One major factor, which is acknowledged but seldom analyzed, is differences in the recognizability of different speakers. In speaker recognition systems such differences are characterized by the use of animal names for different types of speakers, including sheep, goats, lambs and wolves, depending on their behavior with respect to automatic recognition systems. In this paper we propose statistical tests for the existance of these animals and hunt for such animals using results from the 1998 NIST speaker recognition evaluation.
SL980608.PDF (From Author)
SL980608.PDF (Rasterized)
TOP
Progress in Speaker Recognition at Dragon Systems
Authors:
Andres Corrada-Emmanuel, Dragon Systems, Inc. (USA)
Michael Newman, Dragon Systems, Inc. (USA)
Barbara Peskin, Dragon Systems, Inc. (USA)
Lawrence Gillick, Dragon Systems, Inc. (USA)
Robert Roth, Dragon Systems, Inc. (USA)
Page (NA) Paper number 1017
Abstract:
We present a new algorithm for speaker recognition (the Sequential Non-Parametric system, or SNP) that has the potential to overcome two limitations of the current approaches. It uses sequences of frames instead of one frame at a time; and it avoids the need to model a speaker with mixtures of Gaussians by scoring the data non-parametrically. Although at an early stage in its development, SNP's output can be interpolated with that of our GMM system to outperform state-of-the-art GMM's. Comparative results are presented for the 1998 NIST Speaker Recognition Evaluation test set.
SL981017.PDF (Scanned)
TOP
A Comparative Study Of Speaker Verification Systems Using The Polycost Database
Authors:
Tomas Nordström, Telia Research AB (Sweden)
Haakan Melin, KTH, Dept. of Speech, Music and Hearing (Sweden)
Johan Lindberg, KTH, Dept. of Speech, Music and Hearing (Sweden)
Page (NA) Paper number 773
Abstract:
This paper reports on a comparative study of several automatic speaker verification systems using the Polycost database. Polycost is a multi-lingual database with non-native English and mother-tongue speech by subjects from 14 countries. We present results for the first three baseline experiments defined for the database as well as explore the multi-lingual aspects of Polycost in a number of experiments where we compare cross-language and same-language impostor attempts. Our results then lead us to suggest a revised set of baseline experiments.