Speech Duration Research Articles

Enhancing analysis of diadochokinetic speech using deep neural networks

Diadochokinetic speech tasks (DDK) involve the repetitive production of consonant-vowel syllables. These tasks are useful in detecting impairments, differential diagnosis, and monitoring progress in speech-motor impairments. However, manual analysis of those tasks is time-consuming, subjective, and provides only a rough picture of speech. This paper presents several deep neural network models working on the raw waveform for the automatic segmentation of stop consonants and vowels from unannotated and untranscribed speech. A deep encoder serves as a features extractor module, replacing conventional signal processing features. In this context, diverse deep learning architectures, such as convolutional neural networks (CNNs) and large self-supervised models like HuBERT, are applied for the extraction process. A decoder model uses derived embeddings to identify frame types. Consequently, the paper studies diverse deep architectures, ranging from linear layers, LSTM, CNN, and transformers. These architectures are assessed for their ability to detect speech rate, sound duration, and boundary locations on a dataset of healthy individuals and an unseen dataset of older individuals with Parkinson’s Disease. The results reveal that an LSTM model performs better than all other models on both datasets and is comparable to trained human annotators.

250. CHANGES IN AND CLINICAL UTILITY OF MAXIMUM PHONATION TIME AND REPETITIVE SALIVA SWALLOWING TEST SCORES AFTER ESOPHAGECTOMY

Abstract Background Postoperative pneumonia in patients with esophageal cancer occurs due to swallowing dysfunction and aspiration. Recently, intervention by speech-language pathologists (STs) has reported to be important for the prevention of postoperative pneumonia. In this study, we report on the perioperative swallowing rehabilitation led by STs at our hospital. In addition, we focused maximum phonation time (MPT) assessment and repetitive saliva swallowing test (RSST) as swallowing function assessment methods that can identify patients as high risk for pneumonia. We aimed to evaluate the clinical utility of MPT assessment and RSST in patients undergoing oncological esophagectomy. Methods ST intervention was started at the induction of preoperative chemotherapy, and pre-surgical swallowing rehabilitation including oral care, incentive spirometry, abdominal breathing, and breath-hold swallowing was started and continued after surgery. MPT is the longest speech duration after maximal inspiration, and RSST is the number of empty swallowing that can be repeated in 30 seconds. In total, 47 consecutive patients who underwent esophagectomy for esophageal cancer between 2020 and 2023 were eligible. The perioperative changes in MPTs and RSST scores were examined. In addition, univariate and multivariate analyses were performed to identify the predictive factors of postoperative pneumonia. Results The median MPTs before surgery and on postoperative days (PODs) 3, 6, and 10 were 18.4, 7.2, 10.6, and 12.4 s, respectively; postoperative MPTs were significantly lower than preoperative MPT. Meanwhile, there were no significant changes in perioperative RSST scores. A short MPT on POD 6 was one of the independent predictive factors for the incidence of postoperative pneumonia (OR: 12.6, 95%CI: 1.29–123, P = 0.03) in the multivariate analysis. Regarding the onset of postoperative pneumonia, the MPT on POD6 was significantly associated with late-onset pneumonia, but not with early pneumonia (P = 0.02 and P = 0.25, respectively). Conclusion The MPT significantly decreased after esophagectomy. However, the RSST score did not. The MPT on POD6 can be a predictor of postoperative pneumonia. Early resumption of oral intake after esophagectomy might be a risk factor of aspiration pneumonia since expectoration function is not restored, especially in patients with a low MPT on POD 6.

Iterative alignment discovery of speech-associated neural activity.

Objective. Brain-computer interfaces (BCIs) have the potential to preserve or restore speech in patients with neurological disorders that weaken the muscles involved in speech production. However, successful training of low-latency speech synthesis and recognition models requires alignment of neural activity with intended phonetic or acoustic output with high temporal precision. This is particularly challenging in patients who cannot produce audible speech, as ground truth with which to pinpoint neural activity synchronized with speech is not available.Approach. In this study, we present a new iterative algorithm for neural voice activity detection (nVAD) called iterative alignment discovery dynamic time warping (IAD-DTW) that integrates DTW into the loss function of a deep neural network (DNN). The algorithm is designed to discover the alignment between a patient's electrocorticographic (ECoG) neural responses and their attempts to speak during collection of data for training BCI decoders for speech synthesis and recognition.Main results. To demonstrate the effectiveness of the algorithm, we tested its accuracy in predicting the onset and duration of acoustic signals produced by able-bodied patients with intact speech undergoing short-term diagnostic ECoG recordings for epilepsy surgery. We simulated a lack of ground truth by randomly perturbing the temporal correspondence between neural activity and an initial single estimate for all speech onsets and durations. We examined the model's ability to overcome these perturbations to estimate ground truth. IAD-DTW showed no notable degradation (<1% absolute decrease in accuracy) in performance in these simulations, even in the case of maximal misalignments between speech and silence.Significance. IAD-DTW is computationally inexpensive and can be easily integrated into existing DNN-based nVAD approaches, as it pertains only to the final loss computation. This approach makes it possible to train speech BCI algorithms using ECoG data from patients who are unable to produce audible speech, including those with Locked-In Syndrome.

Natural Language Processing Applied to Spontaneous Recall of Famous Faces Reveals Memory Dysfunction in Temporal Lobe Epilepsy Patients.

Epilepsy patients rank memory problems as their most significant cognitive comorbidity. Current clinical assessments are laborious to administer and score and may not always detect subtle memory decline. The Famous Faces Task (FF) has robustly demonstrated that left temporal lobe epilepsy (LTLE) patients remember fewer names and biographical details compared to right TLE (RTLE) patients and healthy controls (HCs). We adapted the FF task to capture subjects' entire spontaneous spoken recall, then scored responses using manual and natural language processing (NLP) methods. We expected to replicate previous group level differences using spontaneous speech and semi-automated analysis. Seventy-three (N=73) adults (28 LTLE, 18 RTLE, and 27 HCs) were included in a case-control prospective study design. Twenty FF in politics, sports, and entertainment (active 2008-2017) were shown to subjects, who were asked if they could recognize and spontaneously recall as much biographical detail as possible. We created human-generated and automatically-generated keyword dictionaries for each celebrity, based on a randomly selected training set of half of the HC transcripts. To control for speech output, we measured the speech duration, total word count and content word count for the FF task and a Cookie Theft Control Task (CTT), in which subjects were merely asked to describe a visual scene. Subjects' responses to FF and CTT tasks were recorded, transcribed, and analyzed in a blinded manner with a combination of manual and automated NLP approaches. Famous face recognition accuracy was similar between groups. LTLE patients recalled fewer biographical details compared to HCs and RTLEs using both the gold-standard human-generated dictionary (24%±12% vs. 31%±12% and 30%±12%, p=0.007) and the automated dictionary (24%±12% vs. 31%±12% and 32%±13%, p=0.007). There were no group level differences in speech duration, total word count, or content word count for either the FF and CTT to explain difference in recall performance. There was a positive, statistically significant relationship between MOCA score and FF recall performance as scored by the human-generated (ρ= .327, p= .029) and automatically-generated dictionaries (ρ= .422, p= .004) for TLE subjects, but not HCs, an effect that was driven by LTLE subjects. LTLE patients remember fewer details of famous people than HCs or RTLE patients, as discovered by NLP analysis of spontaneous recall. Decreased biographical memory was not due to decreased speech output and correlated with lower MOCA scores. NLP analysis of spontaneous recall can detect memory dysfunction in clinical populations in a semi-automated, objective, and sensitive manner.

Prosody in narratives: An exploratory study with children with sex chromosomes trisomies

Although language delays are common in children with sex chromosome trisomies [SCT], no studies have analysed their prosodic abilities. Considering the importance of prosody in communication, this exploratory study aims to analyse the prosodic features of the narratives of 4-year-old children with SCT.Participants included 22 children with SCT and 22 typically developing [TD] children. The Narrative Competence Task was administered to elicit the child's narrative. Each utterance was prosodically analysed considering pitch and timing variables.Considering pitch, the only difference was the number of movements since the utterances of children with SCT were characterised by a lower speech modulation. However, considering the timing variables, children with SCT produced a faster speech rate and a shorter final syllable duration than TD children.Since both speech modulation and duration measures have important syntactic and pragmatic functions, further investigations should deeply analyse the prosodic skills of children with SCT in interaction with syntax and pragmatics.

Enhancing English Pronunciation Assessment in Computer-Assisted Language Learning for College Students

The evolution of computer science and the innovations in language teaching methodologies have paved the way for computer-assisted language learning (CALL) technology to tackle pertinent challenges. While existing CALL systems primarily emphasize vocabulary and grammar acquisition, their evaluation mechanisms often rely on a limited set of criteria, resulting in a simplistic assessment of learners’ pronunciation skills. This oversight underscores the need for a more comprehensive approach. In response, this study targets Chinese college students’ English oral proficiency and aims to enhance the conventional computerized evaluation method. Our approach involves integrating multiple assessment parameters, including pitch, speed, rhythm, and intonation. For instance, pitch assessment is grounded on frequency central feature parameters, while speech speed evaluation considers speech duration, thus enriching the evaluation framework. Through experimental validation, the efficacy of our method in evaluating pitch, speed, rhythm, and intonation has been substantiated, reaffirming its reliability.

The Role of Prosody and Intonation in English Phonology: Implications for Speech Perception and Production

The role of prosody and intonation in English phonology has significant implications for speech perception and production. These phonological processes influence how words are understood and articulated in English. Recent studies have shown that prosody and intonation play a crucial role in verbal communication. This article aims to investigate the importance of prosody and intonation in understanding and producing speech in English language learning intonation refers to the melodic pattern in speech that can change the meaning of a sentence. It is often used to convey emotions or to emphasize a message. On the other hand, prosody encompasses the patterns of pitch, stress, and duration in speech that impact word articulation and comprehension. Both work together to create accurate articulation and understanding of words in English this linguistic study underscores the importance of prosody and intonation in English language learning and elucidates how these two aspects influence speech perception and production. With a deep understanding of prosody and intonation, learners can enhance their ability to communicate in English more effectively and accurately.

Auditory Event-related Potentials for Word Stimuli in Kannada Language Among Native Kannada Speakers with Dementia

Background: The P300 is a positive auditory event-related potential appearing around 300 msec post-oddball stimulus onset. Tone and monosyllabic stimuli have been widely used in P300 research, providing valuable insights into auditory perception, phoneme discrimination, language processing, and other cognitive processes. Bi-syllabic minimal pairs may be more challenging for some individuals because they require processing and discrimination of more complex phonemic structures. The aim of the study was to elicit and analyze the auditory event-related potentials for bi-syllabic word stimuli in the Kannada language among individuals with dementia. Method: The prospective study was conducted at the Department of Speech Pathology and Audiology. Ten native Kannada language speakers (five male; and five female) diagnosed with dementia, aged 55–70 years, were the participants. The naturally produced and recorded CVCV (Consonant vowel consonant vowel) tokens (/ko:ti/ and /ko:thi/ pair; /a: me/ and /a:ne pair) in Kannada were used in the auditory oddball paradigm. Results: The waveforms with positive peak for word (PPW) with clear morphology were used to obtain the latency and amplitude. Mean and Standard deviation (SD) value for the latency of the PPW was 545 ± 38 msec and the amplitude value of PPW was 2.5 ± 1.5 µV. Conclusion: The present study showed that when lengthier speech stimulus duration was used, the latency value of auditory event-related potential was higher. The study also showed that PPW is a reliable peak that could be used in the clinical population.

Automatic speaking valve in tracheo-esophageal speech: treatment proposal for a widespread usage.

Aim of this study was to calculate the percentage of the Automatic Speaking Valve (ASV) use in a large cohort of laryngectomized patients with voice prosthesis (VP) and to analyze the main reasons for non-use. Subsequently, a specific rehabilitation training was proposed. One hundred-ten laryngectomized patients with VP were enrolled in the first phase of the study (census). Among them, 57 patients were included in the second phase (intervention), in which a training based on moving phonatory exercises was proposed. Structured questionnaires were used before and after training in order to investigate ASV use rate (days/week and hours/day; reasons for impeding the ASV use), average adhesive life-time during ASV use; hands-free speech duration; skin irritation. Patients also expressed their degree of on a VAS scale from 0 to 100. In the census phase the percentage of use of ASV (everyday, without problems) was equal to 17.27% (19/110 patients). The main causes of disuse concerned excessive fatigue and poor durability of the adhesives. The analysis of the results pre vs. post-training showed a statistically significant increase (p < 0.05) in all the investigated parameters. Patients reported a good level of treatment compliance (average frequency of performing exercises equal to 4.2 ± 2.5days/week for 1.4 ± 1.01h/day) and high degrees of satisfaction. After treatment, the percentage of use of AVS increased by 43% reaching a rate of 60% (66/110 patients). A specific and targeted approach that simulate the phonatory and breathing difficulties of everyday life can increase the ASV usage rate.

Whisper40: A Multi-Person Chinese Whisper Speaker Recognition Dataset Containing Same-Text Neutral Speech

Whisper speaker recognition (WSR) has received extensive attention from researchers in recent years, and it plays an important role in medical, judicial, and other fields. Among them, the establishment of a whisper dataset is very important for the study of WSR. However, the existing whisper dataset suffers from the problems of a small number of speakers, short speech duration, and lack of neutral speech with the same-text as the whispered speech in the same dataset. To address this issue, we present Whisper40, a multi-person Chinese WSR dataset containing same-text neutral speech spanning around 655.90 min sourced from volunteers. In addition, we use the current state-of-the-art speaker recognition model to build a WSR baseline system and combine the idea of transfer learning for pre-training the speaker recognition model using neutral speech datasets and transfer the empirical knowledge of specific network layers to the WSR system. The Whisper40 and CHAINs datasets are then used to fine-tune the model with transferred specific layers. The experimental results show that the Whisper40 dataset is practical, and the time delay neural network (TDNN) model performs well in both the same/cross-scene experiments. The equal error rate (EER) of Chinese WSR after transfer learning is reduced by 27.62% in comparison.

Linguistic changes in neurodegenerative diseases relate to clinical symptoms.

The detection and characterization of speech changes may help in the identification and monitoring of neurodegenerative diseases. However, there is limited research validating the relationship between speech changes and clinical symptoms across a wide range of neurodegenerative diseases. We analyzed speech recordings from 109 patients who were diagnosed with various neurodegenerative diseases, including Alzheimer's disease, Frontotemporal Dementia, and Vascular Cognitive Impairment, in a cognitive neurology memory clinic. Speech recordings of an open-ended picture description task were processed using the Winterlight speech analysis platform which generates >500 speech features, including the acoustics of speech and linguistic properties of spoken language. We investigated the relationship between the speech features and clinical assessments including the Mini Mental State Examination (MMSE), Mattis Dementia Rating Scale (DRS), Western Aphasia Battery (WAB), and Boston Naming Task (BNT) in a heterogeneous patient population. Linguistic features including lexical and syntactic features were significantly correlated with clinical assessments in patients, across diagnoses. Lower MMSE and DRS scores were associated with the use of shorter words and fewer prepositional phrases. Increased impairment on WAB and BNT was correlated with the use of fewer nouns but more pronouns. Patients also differed from healthy adults as their speech duration was significantly shorter with more pauses. Linguistic changes such as the use of simpler vocabularies and syntax were detectable in patients with different neurodegenerative diseases and correlated with cognitive decline. Speech has the potential to be a sensitive measure for detecting cognitive impairments across various neurodegenerative diseases.

Secondary language impairment in posterior cortical atrophy: insights from sentence repetition.

Posterior cortical atrophy (PCA) is a neurodegenerative syndrome characterized by progressive impairment in visuospatial and perceptual function linked to atrophy of the occipito-parietal cortex. Besides the salient visual impairment, several studies have documented subtle changes in language may also be present. Sentence repetition is a highly constrained linguistic task involving multiple linguistic and cognitive processes and have been shown to be impaired in other AD spectrum disorders, with little consensus on its relevance in PCA. This aim of this study was to further delineate the linguistic and cognitive features of impaired language in PCA using a sentence repetition task. Seven PCA patients and 16 healthy controls verbally repeated 16 sentences from the Boston Diagnostic Aphasia Examination. Responses were transcribed orthographically and coded for accuracy (percentage accuracy; percentage Correct Information Units; Levenshtein Distance) and for temporal characteristics (preparation duration (ms); utterance duration (ms); silent pause duration (ms); speech duration (ms); dysfluency duration (ms)). The potential modulating effects of attentional control and working memory capacity were explored. PCA patients showed lower overall accuracy with retained semantic content of the sentences, and lower phonological accuracy. Temporal measures revealed longer preparation and utterance duration for PCA patients compared to controls, alongside longer speech duration but comparable dysfluency duration. PCA patients also showed comparable silent pause duration to controls. Attentional control, measured using the Hayling sentence completion task, predicted accuracy of sentence repetition. The findings suggest that sentence repetition is impaired in PCA and is characterized by phonological, response planning and execution difficulties, underpinned in part by attentional control mechanisms. The emerging profile of language impairment in PCA suggests vulnerability of similar cognitive systems to other Alzheimer's syndromes, with subtle differences in clinical presentation.

Speech Fluency Production and Perception in L1 (Slovak) and L2 (English) Read Speech.

Research on fluency in native (L1) and non-native (L2) speech production and perception helps us understand how individual L1 speaking style might affect perceived L2 fluency and how this relationship might be reflected in L1 versus L2 oral assessment. While the relationship between production and perception of fluency in spontaneous speech has been studied, the information provided by reading has been overlooked. We argue that reading provides a direct and controlled way to assess language proficiency that might complement information gained from spontaneous speaking. This work analyzes the relationship between speech fluency production and perception in passages of L1 (Slovak) and L2 (English) read by 57 undergraduate Slovak students of English and rated for fluency by 15 English teachers who are Slovak natives. We compare acoustic production measures between L1 and L2 and analyze how their effect on perceived fluency differs for the two languages. Our main finding is that the articulation rate, the overall number of pauses, and the number of between-clause and mid-clause pauses predict ratings differently in L1 Slovak versus L2 English. The speech rate and durations of pauses predict ratings similarly in both languages. The contribution of our results to understanding fluency aspects of spontaneous and read speech, the relationship between L1 and L2, the relationship between production and perception, and to the teaching of L2 English are discussed.

Preceding word information for predicting speech errors in English as foreign language speech

Speech errors, including disfluency errors (e.g., filled pauses (“uh”, “um”), repetition (“I me-mean right now.”)), and mispronunciation of speech segments (e.g., “think” as /sɪŋk/) are natural occurrences in speech production and they can affect speech fluency and proficiency. Detecting these errors is important, especially in assessing second language (L2) learners. Non-native speakers often produce speech errors, even in read speech, due to increased cognitive load when simultaneously producing the current word and processing the upcoming word. By analyzing two L2 speech corpora having different types of errors (disfluency and mispronunciation), the study aims to assess the effectiveness of incorporating preceding word information in speech error prediction. The results indicate that longer syllable duration in the word preceding the mispronunciation error correlates with error production. The study extends its investigation to identify acoustic and lexical features influenced by the presence of speech errors. The study found that speech errors increase duration of speech. The study also found that words with more syllables may cause L2 speakers to produce speech errors.

Speech analysis of teaching assistant interventions in small group collaborative problem solving with undergraduate engineering students

AbstractThis descriptive study focuses on using voice activity detection (VAD) algorithms to extract student speech data in order to better understand the collaboration of small group work and the impact of teaching assistant (TA) interventions in undergraduate engineering discussion sections. Audio data were recorded from individual students wearing head‐mounted noise‐cancelling microphones. Video data of each student group were manually coded for collaborative behaviours (eg, group task relatedness, group verbal interaction and group talk content) of students and TA–student interactions. The analysis includes information about the turn taking, overall speech duration patterns and amounts of overlapping speech observed both when TAs were intervening with groups and when they were not. We found that TAs very rarely provided explicit support regarding collaboration. Key speech metrics, such as amount of turn overlap and maximum turn duration, revealed important information about the nature of student small group discussions and TA interventions. TA interactions during small group collaboration are complex and require nuanced treatments when considering the design of supportive tools. Practitioner notesWhat is already known about this topic Student turn taking can provide information about the nature of student discussions and collaboration. Real classroom audio data of small groups typically have lots of background noise and present challenges for audio analysis. TAs have little training in how to productively intervene with students about collaborative skills. What this paper adds TA interaction with groups primarily focused on task progress and understanding of concepts with negligible explicit support on building collaborative skills. TAs intervened with the groups often which gave the students little time for uptake of their suggestions or deeper discussion. Student turn overlap was higher without the presence of TAs. Maximum turn duration can be an important real‐time turn metric to identify the least verbally active student participant in a group. Implications for practice and/or policy TA training should include information about how to monitor groups for collaborative behaviours and when and how they should intervene to provide feedback and support. TA feedback systems should keep track of previous interventions by TAs (especially in contexts where there are multiple TAs facilitating) and the duration since previous intervention to ensure that TAs do not intervene with a group too frequently with little time for student uptake. Maximum turn duration could be used as a real‐time metric to identify the least verbally active student in a group so that support could be provided to them by the TAs.

Speech detection models for effective communicable disease risk assessment in air travel environments

In environments characterized by elevated noise levels, such as airports or aircraft cabins, travelers often find themselves involuntarily speaking loudly and drawing closer to one another in an effort to enhance communication and speech intelligibility. Unfortunately, this unintentional behaviour increases the risk of respiratory particles dispersion, potentially carrying infectious agents like bacteria which makes the contagion control more challenging. The accurate characterization of the risk associated to speaking, in such a challenging noise environment with multiple overlapping speech sources, is therefore of outmost importance. Among the most advanced signal processing strategies that can be used to accurately determine who spoke when and with whom and for how long but most importantly how loudly, at one location, artificial intelligence-based speaker diarization approaches were considered and adapted for this task. This article details the implementation of speaker diarization algorithms, customized to extract speaker and speech parameters discreetly. Validation and preliminary study results are also provided. The algorithms calculate speech duration and sound pressure level for each sentence and speaker, aiding in assessing viral contaminant spread. The paper focuses on applying these algorithms in noisy environments, particularly in confined spaces with multiple speakers. The findings contribute to proactive measures for containing and managing communicable diseases in air travel settings.

Developing children's ASR system under low-resource conditions using end-to-end architecture

The work presented in this paper aims at enhancing the performance of end-to-end (E2E) speech recognition task for children's speech under low resource conditions. For majority of the languages, there is hardly any speech data from child speakers. Furthermore, even the available children's speech corpora are limited in terms of the number of hours of data. On the other hand, large amounts of adults' speech data are freely available for research as well as commercial purposes. As a consequence, developing an effective E2E automatic speech recognition (ASR) system for children becomes a very challenging task. One may develop an ASR system using adults' speech and then use it to transcribe children's data, but this leads to very poor recognition rates due to the stark differences in the acoustic attributes of adults' and children's speech. In order to overcome these hurdles and to develop a robust children's ASR system employing E2E architecture, we have resorted to several out-of-domain and in-domain data augmentation techniques. For out-of-domain data augmentation, we have explicitly modified adults' speech to render it acoustically similar to that of children's speech before pooling into training. On the other hand, in the case of in-domain data augmentation, we have slightly modified the pitch and duration of children's speech in order to create more data capturing greater diversity. Data augmentation approaches helps in mitigating the ill-effects resulting from the scarcity of data from child domain to a certain extent. This, in turn, reduces the error rates by a large margin. In addition to data augmentation, we have also studied the efficacy of Gamma-tone frequency cepstral coefficients (GFCC) and frequency domain linear prediction (FDLP) technique along with the most commonly used Mel-frequency cepstral coefficients (MFCC) for front-end speech parameterization. Both MFCC as well as GFCC capture and model the spectral envelope of speech. On the other hand, application of linear prediction on the frequency domain representation of speech signal helps to effectively capture the temporal envelope during front-end feature extraction. Employing FDLP features that model the temporal envelope provides important cues for the perception and understanding of stop bursts and, at times, complete phonemes. This motivated us to perform a comparative experimental study of the effectiveness of the three aforementioned front-end acoustic features. In our experimental explorations, the use of proposed data augmentation in combination of FDLP features has shown a relative improvement in character error rate by 67.6% over the baseline system. The combination of data augmentation with MFCC or GFCC features is observed to result in lower recognition performances.

Comparing Machine Learning Models to Determine the Effect of Speech Duration on Speaker Identification within Kazakh Speech Corpus

This paper conducts comparative analysis of speaker identification techniques, namely Gaussian Mixture Models (GMM), Bidirectional Long Short-Term Memory (BiLSTM) neural networks, and Recurrent Neural Networks (RNN), applied to the Kazakh Speech Corpus. The study's central objective is to explore the influence of speech duration on the efficacy of speaker identification. Our results indicate that all models exhibit enhanced accuracy with increased speech duration. Furthermore, the incorporation of delta and fundamental frequency (F0) as supplementary features bolsters the performance of these models. This research contributes to the understanding of how speech duration and additional features can optimize speaker identification techniques, providing valuable insights for future developments in the field.

BEYOND WORDS: HARNESSING SPEECH SOUND FOR SPEAKER AGE AND GENDER DETECTION USING 1D CNN ARCHITECTURE WITH SELF-ATTENTION MECHANISM

Beyond the immediate content of speech, the voice can provide rich information about a speaker's demographics, including age and gender. Estimating a speaker's age and gender offers a wide range of applications, spanning from voice forensic analysis to personalized advertising, healthcare monitoring, and human-computer interaction. However, pinpointing precise age remains intricate due to age ambiguity. Specifically, utterances from individuals at adjacent ages are frequently indistinguishable. Addressing this, we propose a novel, end-to-end approach that deploys Mozilla's Common Voice dataset to transform raw audio into high-quality feature representations using Wav2Vec2.0 embeddings. These are then channeled into our self-attention-based convolutional neural network (CNN) model. To address age ambiguity, we evaluate the effects of different loss functions such as focal loss and Kullback-Leibler (KL) divergence loss. Additionally, we evaluate the accuracy of the estimation at different durations of speech. Experimental results from the Common Voice dataset underscore the efficacy of our approach, showcasing an accuracy of 87% for male speakers, 91% for female speakers and 89% overall accuracy, and an accuracy of 99.1% for gender prediction.

Virtual Reality on Public Speaking Phobia mitigation

The capacity to speak in public requires personal skills that not everyone has. Anxiety and nervousness, many times easily classified as phobia, can represent serious handicaps and committing the expected speaking performance. Being a subject rather explored by contemporaneous literature, this research wants to contribute to exploring Virtual Reality and Gamification strategies for mitigating those difficulties. During this project a VR application for Google Cardboards was developed, to allow speakers to immersively face simulated 3D scenarios. With the aim of representing real-life scenarios as closely as possible, after each speaking session, the assessment process allows the user to find his limitations and reorient future experiences. Speakers are assessed on four indicators, namely voice tone, audience visual contact, head movements, and speech duration. Scenarios and speeches can be created dynamically by specialists, and classified into different difficulty levels, allowing each participant to define his own learning path. Experimentations were done with two distinct groups of students and the main results outcomes are described and analyzed.This research constitutes part of an ERASMUS+ funding project.