Robust Speech Processing in Adverse Environments 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
Linear and Nonlinear Speech Feature Analysis for Stress Classification
Authors:
Guojun Zhou, Robust Speech Processing Lab; Duke Univ. (USA)
John H.L. Hansen, Robust Speech Processing Lab; Duke Univ. (USA)
James F. Kaiser, Robust Speech Processing Lab; Duke Univ. (USA)
Page (NA) Paper number 840
Abstract:
Many stressful environments can deteriorate the performance of speech recognition systems such as aircraft cockpits or high workload task stress/emotional situations. To address this, we investigate a number of linear and nonlinear features and processing methods for stressed speech classification. The linear features include properties of pitch, duration, intensity, glottal source, and the vocal tract spectrum. Nonlinear processing is based on our newly proposed Teager Energy Operator speech feature which incorporates frequency domain critical band filters and properties of the resulting TEO autocorrelation envelope. In this study, we employ a Bayesian hypothesis testing and a hidden Markov model processor as classification methods. Evaluations focused on speech under loud, angry, and the Lombard effect from the SUSAS database. Results using ROC curves and EER based detection show that pitch is the best of the five linear features for stress classification; while the new nonlinear TEO-based feature outperforms the best linear feature by +5.2%, with a reduction in classification rate variability from 8.66 to 3.90.
SL980840.PDF (From Author)
SL980840.PDF (Rasterized)
TOP
Speech Feature Modeling for Robust Stressed Speech Recognition
Authors:
Sahar E. Bou-Ghazale, Rockwell (USA)
John H.L. Hansen, Robust Speech Processing Lab; Duke Univ. (USA)
Page (NA) Paper number 918
Abstract:
It is well known that the performance of speech recognition algorithms degrade in the presence of adverse environments where a speaker is under stress, emotion, or Lombard effect. This study evaluates the effectiveness of traditional features in recognition of speech under stress and formulates new features which are shown to improve stressed speech recognition. The focus is on formulating robust features which are less dependent on the speaking conditions rather than applying compensation or adaptation techniques. The stressed speaking styles considered are simulated angry and loud, Lombard effect speech, and noisy actual stressed speech from the SUSAS database. In addition, this study investigates the immunity of LP and FFT power spectrum to the presence of stress. Our results show that unlike FFT's immunity to noise, the LP power spectrum is more effective than the FFT to stress as well as to a combination of a noisy and stressful environment. Two alternative frequency partitioning methods (M-MFCC, ExpoLog) are proposed and compared with traditional MFCC features for stressed speech recognition. It is shown that the alternate filterbank frequency partitions are more effective for recognition of speech under both simulated and actual stressed conditions.
SL980918.PDF (From Author)
SL980918.PDF (Rasterized)
TOP
Combining Articulatory and Acoustic Information for Speech Recognition in Noisy and Reverberant Environments
Authors:
Katrin Kirchhoff, University of Bielefeld (Germany)
Page (NA) Paper number 873
Abstract:
Robust speech recognition under varying acoustic conditions may be achieved by exploiting multiple sources of information in the speech signal. In addition to an acoustic signal representation, we use an articulatory representation consisting of pseudo-articulatory features as an additional information source. Hybrid ANN/HMM recognizers using either of these representations are evaluated on a continuous numbers recognition task (OGI Numbers95) under clean, reverberant and noisy conditions. An error analysis of preliminary recognition results shows that the different representations produce qualitatively different errors, which suggests a combination of both representations. We investigate various combination possibilities at the phoneme estimation level and show that significant improvements can been achieved under all three acoustic conditions.
SL980873.PDF (From Author)
SL980873.PDF (Rasterized)
TOP
Improving Speaker Identification Performance in Reverberant Conditions using Lip Information
Authors:
Timothy Wark, Speech Research Laboratory, QUT (Australia)
Sridha Sridharan, Speech Research Laboratory, QUT (Australia)
Page (NA) Paper number 294
Abstract:
This paper considers the improvement of speaker identification performance in reverberant conditions using additional lip information. Automatic speaker identification (ASI) using speech characteristics alone can be highly successful, however problems occur with mismatches between training and testing conditions. In particular, we find that ASI performance drops dramatically when given anechoic training but reverberant test speech. Previous work [1][2] has shown that speaker dependant information can be extracted from the static and dynamic qualities of moving lips. Given that lip information is unaffected by reverberation, we choose to fuse this additional information with speech data. We propose a new method for estimating confidence levels to allow adaptive fusion of the audio and visual data. Identification results are presented for increasing levels of artificially reverberated data, where lip information is shown to provide excellent ASI performance improvement.