Speech and Computer: 25th International Conference, SPECOM 2023, Dharwad, India, November 29 – December 2, 2023, Proceedings, Part I (Lecture Notes in Computer Science, 14338)

The two-volume proceedings set LNAI 14338 and 14339 constitutes the refereed proceedings of the 25th International Conference on Speech and Computer, SPECOM 2023, held in Dharwad, India, during November 29–December 2, 2023. The 94 papers included in these proceedings were carefully reviewed and selected from 174 submissions. They focus on all aspects of speech science and technology: ​automatic speech recognition; computational paralinguistics; digital signal processing; speech prosody; natural language processing; child speech processing; speech processing for medicine; industrial speech and language technology; speech technology for under-resourced languages; speech analysis and synthesis; speaker and language identification, verification and diarization. SPECOM 2023 Preface Organization Contents – Part I Contents – Part II Automatic Speech Recognition Extreme Learning Layer: A Boost for Spoken Digit Recognition with Spiking Neural Networks 1 Introduction 2 Methods 2.1 Proposed DELSNN Model 2.2 Speech Signal Encoding 2.3 Neuron Model 2.4 Output Layer Training Through Entropy Minimization 3 Results and Discussion 3.1 Database and Experiments 3.2 Design Decisions and Hyperparameter Tuning 3.3 Performance Metrics 3.4 Effect of Including ELL 3.5 Noise Robustness Evaluation 4 Conclusions and Future Work References EMO-AVSR: Two-Level Approach for Audio-Visual Emotional Speech Recognition 1 Introduction 2 Related Work 2.1 Emotion Recognition 2.2 Audio-Visual Speech Recognition 2.3 Multimodal Corpora 3 Proposed Approach 3.1 Data Pre-processing 3.2 Visual Emotion Recognition Model 3.3 Visual Speech Recognition Model 3.4 Audio Speech Recognition Model 3.5 Cross-Modal Attention Fusion 4 Experimental Evaluation 5 Conclusion References Significance of Audio Quality in Speech-to-Text Translation Systems 1 Introduction 2 Experimental Methodology 2.1 Dataset Description and Pre-processing 2.2 Building of Models 2.3 Inference and Evaluation Methods 3 Results and Analysis 3.1 Performance Analysis Using High-Quality Audios 4 Conclusion and Future Work References Everyday Conversations: A Comparative Study of Expert Transcriptions and ASR Outputs at a Lexical Level 1 Introduction 2 Data and Method 2.1 Data 2.2 Method 3 Word Lists Statistics 3.1 Comparison of Frequency Lists 3.2 Words that Are Least Recognized 4 Words Not Found in the Expert Transcription 5 Conclusions References Improving Automatic Speech Recognition with Dialect-Specific Language Models 1 Introduction 2 Dataset 3 Methodology 3.1 Acoustic Model 3.2 Language Model 4 Experiments Setup 5 Results 6 Analysis and Discussion 7 Conclusions References Emotional Speech Recognition of Holocaust Survivors with Deep Neural Network Models for Russian Language 1 Introduction 2 Related Work 3 Data Overview 4 Data Collection Pipeline 4.1 Data Analysis 4.2 Data Preprocessing 5 Speech Recognition Results 6 Conclusion References Computational Paralinguistics Aggregation Strategies of Wav2vec 2.0 Embeddings for Computational Paralinguistic Tasks 1 Introduction 2 Proposed Methods 2.1 Wav2vec 2.0 Embeddings 2.2 Embedding Aggregation 3 Databases 3.1 The iHEARu-EAT Database 3.2 The URTIC Database 3.3 The AIBO Database 4 Experimental Setup 4.1 Wav2vec 2.0 Embeddings 4.2 Embedding Aggregation 4.3 Classification and Evaluation 5 Experimental Results 5.1 Feature Combinations 6 Conclusion and Future Work References Rhythm Formant Analysis for Automatic Depression Classification 1 Introduction 2 Rhythm Formant Analysis 3 Classification System Using FM Rhythm Formants 4 Dataset and Experimental Description 5 Evaluation Metrics and Results 6 Conclusion and Future Work References Determining Alcohol Intoxication Based on Speech and Neural Networks 1 Introduction 2 Related Works 3 Materials and Methods 3.1 Creation of Dataset 3.2 Creating Input Data for the Neural Network 3.3 Neural Network Training 4 Results 5 Conclusion References Linear Frequency Residual Cepstral Coefficients for Speech Emotion Recognition 1 Introduction 2 Linear Prediction 3 Linear Frequency Residual Cepstral Coefficients 4 Experimental Setup 4.1 Dataset Used 4.2 Classifiers Used 4.3 Baseline Considered 5 Experimental Results 5.1 Spectrographic Analysis 5.2 Impact of LP Order 5.3 Significance of Pitch Contour 5.4 Results with Score-Level Fusion 5.5 Results with Classifier-Level Fusion 5.6 Performance of LFRCC on SER 6 Summary and Conclusion References Enhancing Stutter Detection in Speech Using Zero Time Windowing Cepstral Coefficients and Phase Information 1 Introduction 2 Related Work 3 Methods 3.1 ZTW 3.2 ZTWCC 3.3 Phase Information 4 Proposed Work 5 Experimental Setup 5.1 Database 5.2 Metrics 6 Results and Discussions 7 Conclusion References Source and System-Based Modulation Approach for Fake Speech Detection 1 Introduction 2 Motivation 3 Proposed FSD System 3.1 Residual Modulation Spectrogram 3.2 ResNet-[34] Classifier 4 Experiments and Results 4.1 Database Description 4.2 Baseline Method 4.3 ResNet-[34] Classification Result 5 Discussion 6 Conclusion and Future Work References Digital Signal Processing Investigation of Different Calibration Methods for Deep Speaker Embedding Based Verification Systems 1 Introduction 2 Problem Overview 2.1 Condition-Aware Calibration 3 Embedding Extractor 4 Calibration 4.1 Modifications 5 Experimental Setup 5.1 Training Dataset 5.2 Test Datasets and Metrics 6 Experimental Results and Discussions 7 Conclusions References Learning to Predict Speech Intelligibility from Speech Distortions 1 Introduction 2 Our Approach 2.1 SNR vs ESTOI for Various Distortions 2.2 Models 3 Experimental Details and Results 4 Conclusion and Future Work References Sparse Representation Frameworks for Acoustic Scene Classification 1 Introduction 2 Sparse Representation Frameworks for ASC 2.1 Sparse Representation Classification (SRC) 2.2 Deep Sparse Representation Classification (DSRC) 3 Discriminative Sparse Auto-Encoder (DSAE) Framework 4 Experimental Evaluation 4.1 Dataset Description 4.2 Feature Extraction 4.3 Results and Discussion 5 Conclusion References Driver Speech Detection in Real Driving Scenario 1 Introduction 1.1 Related Works 1.2 Motivation 2 Natural and Real Driving Speech Corpus 3 Characteristics of Driver's Speech 4 Driving Speech Detection 4.1 Features Sets 4.2 Classification Setup 4.3 Performance Analysis 5 Conclusions References Regularization Based Incremental Learning in TCNN for Robust Speech Enhancement Targeting Effective Human Machine Interaction 1 Introduction 2 Proposed Speech Enhancement Framework 3 Regularization Strategy for Incremental Learning 4 Experiments 5 Results 6 Conclusions References Candidate Speech Extraction from Multi-speaker Single-Channel Audio Interviews 1 Introduction 2 Problem Formulation 3 Candidate Speech Identification 4 Experimental Setup and Results 5 Conclusion References Post-processing of Translated Speech by Pole Modification and Residual Enhancement to Improve Perceptual Quality 1 Introduction 2 Direct Speech-to-Speech Translation 2.1 Training Stage 2.2 Inference Stage 3 Speech Enhancement 3.1 Pole Modification 3.2 Residual Enhancement by Weighting 3.3 LP Residual Replacement 4 Experimental Setup, Results and Discussion 4.1 Dataset Description 4.2 Experimental Setup 4.3 Performance Evaluation and Discussion 5 Conclusion and Future Work References Region Normalized Capsule Network Based Generative Adversarial Network for Non-parallel Voice Conversion 1 Introduction 2 Proposed Model 2.1 Region Normalized Technique-Based Generator 2.2 Caps-Net Discriminator 2.3 Loss Functions 3 Experimental Design 3.1 Dataset Description 3.2 Training Details 3.3 Experimental Setup 4 Results and Discussion 4.1 Objective Evaluation 4.2 Subjective Evaluation 5 Conclusion References Speech Enhancement Using LinkNet Architecture 1 Introduction 2 Speech Enhancement 3 Deep Learning Approaches 3.1 Encoder Decoder Architecture - 2DCNN 3.2 Autoencoder with Skip Connections 3.3 UNet 4 Experimental Setup 4.1 Data Preprossessing 4.2 Network Architecture 5 Results 6 Conclusion References ATT:Adversarial Trained Transformer for Speech Enhancement 1 Introduction 2 Adversarial Trained Transformer 2.1 Overall Network Overview 2.2 Generator 2.3 Discriminator 3 Experimental Setup 3.1 Data Set 3.2 ATT Setup 4 Experimental Results 5 Conclusion References Human Identification by Dynamics of Changes in Brain Frequencies Using Artificial Neural Networks 1 Introduction 2 Bioelectric Signals in Process Automation Tasks 3 Collection and Preparation of Experimental Data 4 Results of Deep Machine Learning and Machine Classification 5 Evaluation of the Adequacy of the Classification Models 6 Conclusion References Speech Prosody Analysis of Formant Trajectories of a Speech Signal for the Purpose of Forensic Identification of a Foreign Speaker 1 Introduction 2 Solution of the Task of Identification in Relation to the Romani Language 2.1 Method 2.2 Comparison of Formant Trajectories and Spectrograms of Identical Speech Signals 2.3 Comparison of Formant Trajectories for Speech Signals Recorded Under Different Conditions 3 Results of the Investigation 4 Conclusion References Gestures vs. Prosodic Structure in Laboratory Ironic Speech 1 Introduction 1.1 Gestures Description 1.2 Synchronization with Prosodic Structure 1.3 Paralinguistic Characteristics of Ironic Speech 2 Material and Method 2.1 Perceptual Experiments 3 Results 3.1 Single- and Multichannel Perception of Ironic and Non-ironic Utterances 3.2 Gestures 3.3 Synchronization 4 Conclusion References Sounds of ence: Acoustics of Inhalation in Read Speech 1 Introduction 2 Methodology 2.1 Speech Database and Annotation 2.2 Acoustic and Statistical Analyses 3 Results 3.1 Amount of Inhalation in Speech 3.2 Correlation Between Formants in Inhalation and Contextual Vowels 4 Discussion and Conclusion References Prolongations as Hesitation Phenomena in Spoken Speech in First and Second Language 1 Introduction 2 Material and Methodology 2.1 Prolongations of Vowels and Consonants in Russian 2.2 Prolongations of Vowels and Consonants in Chinese 2.3 Problems of Analysis and Decisions Made 2.4 Phonetic Properties of Prolongations 3 Results 3.1 Quantitative Overview 3.2 Frequency List of Words with Prolongations in Russian and Chinese 3.3 Position of Prolongation 3.4 Prolongations of Vowels and Consonants 3.5 Isolated Prolongations and Prolongations in Hesitation Chain 4 Conclusion References Study of Indian English Pronunciation Variabilities Relative to Received Pronunciation 1 Introduction 2 Data Annotation and Pre-processing 2.1 Indic TIMIT Corpus 2.2 Data Preparation 3 Indian English in Linguistic Literature 3.1 Context Dependent Phonetic Rules 4 Data Analysis 4.1 Procedure 4.2 Discussion 4.3 Efficacy of G2P System Based on Phonetic Rules 5 Conclusion References Multimodal Collaboration in Expository Discourse: Verbal and Nonverbal Moves Alignment 1 Introduction 2 Theoretical Framework 2.1 Verbal and Nonverbal Communicative Moves in Collaborative Communication 2.2 Verbal and Nonverbal Alignment in Collaborative Communication 3 Data and Methods 3.1 Experiment Design and Data 3.2 Methods 4 Results and Discussion 4.1 Communicative Moves Within a Collaborative Act 4.2 Communicative Moves Distribution in Collaborative Expository Discourse 4.3 Multimodal Intentional and Spontaneous Collaboration 5 Final Remarks References Association of Time Domain Features with Oral Cavity Configuration During Vowel Production and Its Application in Vowel Recognition 1 Introduction 2 Methodology 2.1 Brief Description of Bangla Language 2.2 Vowel Production Mechanism and Mapping to Tongue Height and Position 2.3 Time Domain Parameters (TDP) 2.4 Algorithms 3 Experimental Details 3.1 Data Gathering and Preparation 3.2 Experimental Procedure 3.3 Recognition System 4 Result and Discussion 4.1 ER and PA 4.2 PPD and ZCR 4.3 Classification of Vowel 5 Conclusion References Prosodic Interaction Models in a Conversation 1 Introduction 2 Methodology 2.1 Data 2.2 Procedure 2.3 Measurements 3 Results 3.1 Model 1: Convergence – Divergence 3.2 Model 2: Convergence – Intensified Convergence 3.3 Model 3: Divergence – Convergence 4 Conclusions and Discussion References Natural Language Processing Development and Research of Dialogue Agents with Long-Term Memory and Web Search 1 Introduction 2 Related Work 3 Methods 3.1 Overview of Datasets 3.2 Model Architecture 4 Experiments 5 Conclusion References Pre- and Post-Textual Contexts in Assessment of a Message as Offensive or Defensive Aggression Verbalization 1 Introduction 1.1 Types of Aggression 2 Experimental Research on Defensive Aggression Verbalization 2.1 Research Material 2.2 Research Procedure 2.3 Tasks for the Recipients 2.4 Respondents’ Demographic Data 3 Results 3.1 Out-Of-Context Message Assessment 3.2 In-Context Message Assessment 4 Conclusion References Boosting Rule-Based Grapheme-to-Phoneme Conversion with Morphological Segmentation and Syllabification in Bengali 1 Introduction 2 Literature Survey 2.1 Limitations 2.2 Objective 3 Proposed G2P Conversion Approach 3.1 Morphological Segmentation 3.2 Rule-Based G2P Conversion 3.3 Syllabification 3.4 Pronunciation Refining 4 Corpus and Participants 4.1 Corpus 4.2 Participants 5 Experiments, Results, and Findings 5.1 Answering RQ1: Evaluating Proposed G2P Approach 5.2 Answering RQ2: Evaluating Proposed G2P Approach 5.3 Answering RQ3: Evaluating TTS System 5.4 Answering RQ4: Evaluating ASR System 6 Conclusion References Revisiting Assessment of Text Complexity: Lexical and Syntactic Parameters Fluctuations 1 Introduction 2 Literature Review 3 Methods and Data 4 Analysis 5 Conclusion References Analysis of Natural Language Understanding Systems with L2 Learner Specific Synthetic Grammatical Errors Based on Parts-of-Speech 1 Introduction 2 Background 3 Methods 3.1 Synthesizing Grammatical Errors 3.2 Intent Detection and Slot Filling Performance 3.3 POS Analysis 3.4 Attention Analysis 4 Results 4.1 Intent Detection and Slot Filling 4.2 POS Analysis 4.3 Attention Analysis 5 Discussion 6 Conclusions 7 Future Work References On the Most Frequent Sequences of Words in Russian Spoken Everyday Language (Bigrams and Trigrams): An Experience of Classification 1 Introduction 2 Materials and Methodology of N-gram Analysis 3 Analysis of Bigrams in Contemporary Russian Oral Discourse: Frequency Characteristics and Typology of Units Obtained with Quantitative Data 4 Analysis of Trigrams in Contemporary Russian Oral Discourse: Frequency Characteristics and Typology of the Obtained Units with Quantitative Data 5 Conclusion References Child Speech Processing Recognition of the Emotional State of Children by Video and Audio Modalities by Indian and Russian Experts 1 Introduction 2 Methods 2.1 Participants of the Study 2.2 Data Collection 2.3 Dataset 2.4 Data Analysis 3 Results 3.1 Perceptual Experiment 3.2 Characteristics of Video and Audio Fragments Correctly Recognized by Indian and Russian Experts (Probability 0.75–1.0) 4 Discussion and Future Work 5 Conclusion References Effect of Linear Prediction Order to Modify Formant Locations for Children Speech Recognition 1 Introduction 2 Effect of Linear Prediction Order to Modify Formant Location 3 Database and Experimental Setup 4 Results and Discussion 5 Conclusion References Gammatone-Filterbank Based Pitch-Normalized Cepstral Coefficients for Zero-Resource Children's ASR 1 Introduction 2 Proposed Pitch-Normalized Front-End Acoustic Features 3 Experimental Evaluations 3.1 Database and Experimental Specification 3.2 Results and Discussions 4 Conclusion References System Assisted Vocal Response Analysis and Assessment of Autism in Children: A Machine Learning Based Approach 1 Introduction 2 Speech and Vocal Response Based Autism Assessment 2.1 Review of Literature 2.2 Purpose of the Paper 2.3 ISAA Criteria for Vocal Response Assessment 3 Data Collection Methodology 3.1 Stimulus and Approach Planning 3.2 Design of Stimulus 3.3 Vocal Response Capture Framework 3.4 Response Data Collection Software 4 Child Response Annotation, Statistics and Analysis 4.1 Child Response Annotation 4.2 Child Response Statistics 4.3 Observations While Data Collection and Response Analysis 5 Response Modelling Experiments 5.1 Experimental Settings 5.2 Adult Speech vs Child Response Experiment 5.3 Child Response Classification Experiment 5.4 Child Response Classification on Merging Speech and Non-Speech 6 Conclusion References Addressing Effects of Formant Dispersion and Pitch Sensitivity for the Development of Children's KWS System 1 Introduction 2 Proposed Approach for Computation of MFCC by Filtering the Formant Enhanced ST-MS 2.1 Analysis of Proposed Approach for Reducing the Effects of Pitch and Formant Dispersion 3 Experimental Setup 4 Experimental Results and Discussions 4.1 Performance of TASS-MFCC-ARP in Pitch-Matched and Mismatched Test Conditions 4.2 Impact of Mel-Filterbank Size on the Proposed Feature 4.3 Performance Comparison of TASS-MFCC-ARP with Other Features for the Modified Size of Mel-Filterbank 4.4 Performance of TASS-MFCC-ARP with Data-Augmented Training 5 Conclusion References Emotional State of Children with ASD and Intellectual Disabilities: Perceptual Experiment and Automatic Recognition by Video, Audio and Text Modalities 1 Introduction 2 Methods 2.1 Participants of the Study 2.2 Data Collection 2.3 Dataset 2.4 Perceptual Study 2.5 Automatic Analysis of Facial Expression and Emotional Speech of Children 3 Results 3.1 Perceptual Experiment 3.2 Automatic Analysis of Facial Expression 3.3 Automatic Analysis of Child Speech 4 Discussion 5 Conclusion References Linear Frequency Residual Features for Infant Cry Classification 1 Introduction 2 Linear Prediction (LP) Residual 3 Linear Frequency Residual Cepstral Coefficients (LFRCC) 4 Experimental Setup 4.1 Dataset Used 4.2 Classifiers Used 4.3 Baseline Used 5 Experimental Results 5.1 Spectrographic Analysis 5.2 Effect of LP Order 5.3 Results for Matched Conditions 5.4 Results for Mismatched Database 6 Summary and Conclusion References Speech Processing for Medicine Identification of Voice Disorders: A Comparative Study of Machine Learning Algorithms 1 Introduction 2 Related Work 3 Methodology 3.1 Preprocessing 3.2 Feature Extraction 3.3 Handling Imbalanced Data 3.4 Model Building 4 Experiments and Results 4.1 Dataset Description 4.2 Results 5 Conclusion References Transfer Learning Using Whisper for Dysarthric Automatic Speech Recognition 1 Introduction 2 Proposed Work 2.1 Introduction to Whisper 2.2 Weakly Supervised Training of Whisper 2.3 Whisper Models 2.4 Proposed Transfer Learning Methodology 2.5 Working of the Employed Pipeline 3 Experimental Setup 3.1 Dataset Used 3.2 Classification of Isolated Dysarthric Speech 4 Experimental Results 4.1 Severity-Level Independent ASR System 4.2 Severity-Level Dependent ASR System 5 Summary and Conclusions References Significance of Duration Modification in Reducing Listening Effort of Slurred Speech from Patients with Traumatic Brain Injury 1 Introduction 2 Analysis of Slurred Speech 2.1 F0 Analysis 2.2 Speaking Rate 3 Epoch Based Duration Modification of Speech 4 Perceptual Evaluations and Results 4.1 Data for Perceptual Evaluations 4.2 Neurological Diagnosis of the Patients's Cognitive Impairment 4.3 Perceptual Evaluation for Assessing Listening Effort 5 Summary and Conclusion References Speech Signal Segmentation into Silence, Unvoiced and Vocalized Sections in Speech Rehabilitation 1 Introduction 2 Experiment and Results 2.1 The Segmentation Algorithm 2.2 The Data Set 2.3 The Result of the Algorithm with Initial Constant Values 2.4 The Algorithm Improvement Based on Minimization of the Number of Classification Errors 2.5 The Algorithm Improvement Based on Minimizing the Distance to Manual Boundaries 3 Conclusion References Respiratory Sickness Detection from Audio Recordings Using CLIP Models 1 Introduction 2 CLIP Models 3 Dataset Description 4 Proposed Algorithm 5 Simulations and Results 6 Conclusions References Investigating the Effect of Data Impurity on the Detection Performances of Mental Disorders Through Spoken Dialogues 1 Introduction 2 Experimental Details 2.1 Dataset 2.2 Methodology 2.3 Model Architectures 3 Results 3.1 MDD Detection 3.2 PTSD Detection 3.3 Comparison with Similar Works Involving Data Modification 4 Summary and Discussion 5 Conclusions References Author Index This book constitutes the refereed proceedings of the 16th International Conference on Speech and Computer, SPECOM 2014, held in Novi Sad, Serbia. The 56 revised full papers presented together with 3 invited talks were carefully reviewed and selected from 100 initial submissions. It is a conference with long tradition that attracts researchers in the area of computer speech processing (recognition, synthesis, understanding etc.) and related domains (including signal processing, language and text processing, multi-modal speech processing or human-computer interaction for instance).