Speech Samples Research Articles

Maxillectomy patients' speech and performance of contemporary speaker-independent automatic speech recognition platforms in Japanese.

Automatic speech recognition (ASR) can potentially help older adults and people with disabilities reduce their dependence on others and increase their participation in society. However, maxillectomy patients with reduced speech intelligibility may encounter some problems using such technologies. To investigate the accuracy of three commonly used ASR platforms when used by Japanese maxillectomy patients with and without their obturator placed. Speech samples were obtained from 29 maxillectomy patients with and without their obturator and 17 healthy volunteers. The samples were input into three speaker-independent speech recognition platforms and the transcribed text was compared with the original text to calculate the syllable error rate (SER). All participants also completed a conventional speech intelligibility test to grade their speech using Taguchi's method. A comprehensive articulation assessment of patients without their obturator was also performed. Significant differences in SER were observed between healthy and maxillectomy groups. Maxillectomy patients with an obturator showed a significant negative correlation between speech intelligibility scores and SER. However, for those without an obturator, no significant correlations were observed. Furthermore, for maxillectomy patients without an obturator, significant differences were found between syllables grouped by vowels. Syllables containing /i/, /u/ and /e/ exhibited higher error rates compared to those containing /a/ and /o/. Additionally, significant differences were observed when syllables were grouped by consonant place of articulation and manner of articulation. The three platforms performed well for healthy volunteers and maxillectomy patients with their obturator, but the SER for maxillectomy patients without their obturator was high, rendering the platforms unusable. System improvement is needed to increase accuracy for maxillectomy patients.

Unveiling early signs of Parkinson’s disease via a longitudinal analysis of celebrity speech recordings

Numerous studies proposed methods to detect Parkinson’s disease (PD) via speech analysis. However, existing corpora often lack prodromal recordings, have small sample sizes, and lack longitudinal data. Speech samples from celebrities who publicly disclosed their PD diagnosis provide longitudinal data, allowing the creation of a new corpus, ParkCeleb. We collected videos from 40 subjects with PD and 40 controls and analyzed evolving speech features from 10 years before to 20 years after diagnosis. Our longitudinal analysis, focused on 15 subjects with PD and 15 controls, revealed features like pitch variability, pause duration, speech rate, and syllable duration, indicating PD progression. Early dysarthria patterns were detectable in the prodromal phase, with the best classifiers achieving AUCs of 0.72 and 0.75 for data collected ten and five years before diagnosis, respectively, and 0.93 post-diagnosis. This study highlights the potential for early detection methods, aiding treatment response identification and screening in clinical trials.

FPGA-Based Design of a Ready-to-Use and Configurable Soft IP Core for Frame Blocking Time-Sampled Digital Speech Signals †

‘Frame blocking’ or ‘Framing’ is a technique that divides a time-sampled speech or audio signal into consecutive and equi-sized short-time frames, either overlapped or non-overlapped, for analysis. The framing hardware architectures (FHA) in the literature support framing speech or audio samples of specific word size with specific frame size and frame overlap size. However, speech and audio applications often require framing signal samples of varied word sizes with varied frame sizes and frame overlap sizes. Therefore, the existing FHAs must be redesigned appropriately to keep up with the variability in word size, frame size and frame overlap size, as demanded across multiple applications. Redesigning the existing FHAs for each specific application is laborious, prompting the need for a configurable intellectual property (IP) core. The existing FHAs are inappropriate for creating configurable IP cores as they lack adaptability to accommodate variability in frame size and frame overlap size. Therefore, to address these issues, a novel FHA, adaptable to accommodate the desired variability, is proposed. Furthermore, the proposed FHA is transformed into a field-programmable gate array-based soft, ready-to-use and configurable frame blocking IP core using the Xilinx® Vivado™ tool (San Jose, CA, USA). The resulting IP core is versatile, offering configurability for framing in numerous applications incorporating real-time digital speech and audio systems. This research article discusses the proposed FHA and frame blocking IP core in detail.

Improved Speech Authenticity Detection in Chinese–English Bilingual Contexts

The rapid evolution of voice technology has heightened the need for robust detection systems to distinguish between authentic and tampered speech. Recent competitions have significantly advanced the development of countermeasures against spoofing attacks. However, while advancements in detection technologies have been notable, existing methods often focus on a single type of tampering and language. Our contribution lies in developing an improved model that integrates an enhanced ResNet architecture with an LSTM to improve the detection of tampered audio, particularly in challenging multilingual scenarios. In the experiments, we built a hybrid dataset from self-recording Chinese speech and public VCTK2 English samples, enhanced the ResNet model generalization capabilities, and evaluated our approach using the bilingual dataset. Experiment results demonstrate that the proposed approach achieves a superior performance with an equal error rate of 11.62%, even in the face of bilingual conditions, and, more importantly, outperforms the leading models from ASVSpoof 2021 and ADD 2022 competitions. We also employed advanced tampering techniques, including CycleGAN voice conversion and auto splicing, to simulate real-world tampering scenarios and verify the effectiveness of the proposed approach.

The Pronunciation Problems Encounter EFL Learners in Spoken English at University

This research examines the pronunciation difficulties of EFL undergraduate learners at universities when pronouncing English phonemes. It collects data using recorded speech samples, pronunciation tests and questionnaires as part of a research method for collecting data. The instruments used for this study are collecting data and analyzing them. The test materials are prepared in speech levels. Therefore, the observations of these were based on auditory impressions. The results suggest that the participants had a pronunciation challenge with the phonemes in English. This research gives insights and supports ESL teachers in developing teaching techniques that will lessen or eliminate future issues with Arabic speakers' pronunciation of English phonemes.

Enhanced Speech Processing for Air Traffic Control Using an Optimized Generative Adversarial Network

In order to improve the quality of voice calls during air traffic control, an improved SEGAN air traffic control speech enhancement algorithm is proposed. Aiming at the problem that the traditional speech enhancement algorithm based on generative adversarial network (SEGAN) is drowned under low signal-to-noise ratio conditions, a multi-stage, multi-mapping, multi-dimensional output generator and a multi-scale, multi-discriminator network model are proposed based on the SEGAN network model. First, the speech semantic features are extracted based on the deep neural network structure to complete the semantic segmentation of air traffic control speech. Secondly, multiple generators are set to further optimize the speech signal. Then, a downsampling module is added to the convolution layer to improve the model's utilization of speech information and reduce the loss of speech information. Finally, multi-scale, multiple discriminators are used to learn the distribution law and information of speech samples in multiple directions. The results show that under low signal-to-noise ratio conditions, the improved SEGAN model improves the short-term objective intelligibility (STOI) and the perceptual evaluation of speech quality (PESQ) by and respectively , which can quickly and effectively perform air traffic control speech enhancement and provide preparation for subsequent air traffic control speech recognition.

The Relationship between Traditional Acoustic Measures and Cepstral Analysis of Voice.

In this study, the correlations between traditional acoustic measures (TAMs) and cepstral analysis (CA) were explored in Persian. This investigation was a cross-sectional study including 179 dysphonic (n=141) and normophonic (n=38) speakers. The TAMs (jitter, shimmer, and noise-to-harmonic ratio) and CA (cepstral peak prominence, and cepstral peak prominence-smoothed) values were obtained during vowel prolongation, reading a standard sentence, and a non-standard running speech sample using Praat software. The difference of acoustic measures between normophonic and dysphonic speakers and intercorrelation among acoustic measures and correlation between the acoustic measures and perceived dysphonia levels was analyzed with independent t test, Mann-Whitney U test, Pearson, Spearman, and Kendall's Tau-b correlation tests using IBM SPSS Statistics. The findings showed that dysphonic speakers had higher TAM values and lower CA values than normophonic speakers. (P < 0.05). In dysphonic speakers, a large correlation was discovered among all acoustic measurements (r = 0.52 to 0.96; P < 0.05), while in various perceived dysphonic speakers, there was a correlation of varying strength (r = 0.25 to 0.97; P < 0.05). Ultimately, there was a significant small-to-large correlation between the acoustic measures and perceived dysphonia levels (r = 0.34-0.58; P < 0.05). This research demonstrated that Persian speakers with dysphonia experienced a rise in TAM and a corresponding reduction in CA. In the future, multi-parametric indices can be developed using both TAM and CA to include various aspects of vocal production and yield a single, comprehensive value.

DSTM: A transformer-based model with dynamic-static feature fusion in speech emotion recognition

With the support of multi-head attention, the Transformer shows remarkable results in speech emotion recognition. However, existing models still suffer from the inability to accurately locate important regions in semantic information at different time scales. To address this problem, we propose a Transformer-based network model for dynamic-static feature fusion, composed of a locally adaptive multi-head attention module and a global static attention module. The locally dynamic multi-head attention module adapts the attention window sizes and window centers of the different regions through speech samples and learnable parameters, enabling the model to adaptively discover and pay attention to valuable information embedded in speech. The global static attention module enables the model to use each element in the sequence fully and learn critical global feature information by establishing connections over the entire input sequence. We also use the data mixture training method to train our model and introduce the CENTER LOSS function to supervise the training of the model, which can better speed up the fitting speed of the model and alleviate the sample imbalance problem to a certain extent. This method achieved good performance on the IEMOCAP and MELD datasets, proving that our proposed model structure and method have better accuracy and robustness.

Voice pitch and gender in autism.

It is has often been observed that autistic individuals have higher-pitched voices than non-autistic ones, but no clear explanation for this difference has been put forth. However, autistic males are still dramatically over-represented in published research, including the acoustic studies that report higher pitch in autistic participants. In this study, we collected speech samples from a group of autistic and neurotypical adults that, unlike in most studies, was perfectly balanced between groups and genders. In this gender-balanced sample, pitch was significantly higher in autistic versus neurotypical men, but lower in autistic versus neurotypical women. Overall, women tend to have higher-pitched voices than men, but the magnitude of this difference is culture dependent and may be significantly influenced by the internalisation of normative expectations towards one's gender. We propose that higher pitch in autistic males and lower pitch in autistic females could be due, at least in part, to a lesser integration of sociolinguistic markers of gender. Our report shows that speech atypicality should not be operationalised in terms of systematic and unidirectional deviation from the neurotypical baseline.

A Comparative Effectiveness Study of Speech and Surgical Outcomes: Study Overview.

To provide an overview of the Cleft Outcomes Research NETwork (CORNET) and the CORNET Speech and Surgery study. The study is (1) comparing speech outcomes and fistula rate between two common palate repair techniques, straight-line closure with intra-velar veloplasty (IVVP) and Furlow Double-Opposing Z-palatoplasty (Furlow Z-plasty); (2) summarizing practice variation in the utilization of early intervention speech-language (EI-SL) services; and (3) exploring the association between EI-SL services and speech outcomes. Prospective, longitudinal, observational, comparative effectiveness, multi-center. Twenty sites across the United States. One thousand two hundred forty-seven children with cleft palate with or without cleft lip (CP ± L). Children with submucous cleft palate or bilateral sensorineural severe to profound hearing loss were excluded from participation. Straight-line closure with IVVP or Furlow Z-plasty based on each surgeon's standard clinical protocol. The primary study outcome is perceptual ratings of hypernasality judged from speech samples collected at 3 years of age. Secondary outcomes are fistula rate, measures of speech production, and quality of life. The statistical analyses will include generalized estimating equations with propensity score weighting to address potential confounders. Recruitment was completed in February 2023; 80% of children have been retained to date. Five hundred sixty two children have completed their final 3-year speech assessment. Final study activities will end in early 2025. This study addresses long-standing questions related to the effectiveness of the two most common palatoplasty approaches and describes CORNET which provides an infrastructure that will streamline future studies in all areas of cleft care.

Evaluation of the effect of respiratory airflow on captured speech quality: application to pop filters

Recording speech is often a matter of proper experimental design to obtain the best signal quality. This notably includes appropriate microphone positioning and gain adjustment. However, some radio communication devices require that the microphone be placed in a very close proximity to the mouth, therefore capturing the respiratory airflow, as well as the air bursts from plosive consonants. In this paper, we assess an objective evaluation method for the quality of speech signals captured in such adverse conditions. It employs an artificial mouth coupled with a steady airflow to compute the Speech Transmission Index (STI). We apply this method to the evaluation of four commercially available pop filters. The addition of a 30~L/min airflow reduces the STI with the microphone alone from excellent to fair. With soft fabric filters, we gain back the original STI, while it remains very slightly degraded with a metal mesh filter. The objective metric is finally confronted with a listening test. The perceived quality of simulated samples with the added airflow is similar to the STI results. However, no significant difference between the filters is found with real speech samples containing plosive consonants and no respiration noise.

Machine-Based and Auditory Identification of Slavic Languages

This paper presents a comparison of auditory and machine-based identification of linguistic origins. Two studies were conducted to assess the ability of lay listeners and a stateof-the-art machine approach to identify Slavic L1 from delexicalized speech samples. The first study involved 228 native speakers of the four Slavic languages (Bulgarian, Czech, Polish and Russian) who had not received any prior training in Slavic philology, phonetics, linguistics, or forensic science. Their task was to identify the linguistic origins of speakers when exposed to limited phonetic cues. The stimuli consisted of meaningless logatomes to control for the lexical information. The second study employed machine-based identification of a spoken language, based on two distinct approaches: (1) formant structure of phonetic signal and (2) a neural network and vector representation of speech samples. The data showed that Slavic native speakers, even when exposed to limited auditory cues, are able to identify speakers’ L1s. Interestingly, in the context of the Bulgarian language, the machine-based identification method performed better than the lay listeners. The results of the experiments provide insight into the advantages of hybrid approaches in investigations related to LADO (Language Analyses for the Determination of Origin). Furthermore, the outcomes of this comparison may contribute to the debate on the involvement of native speakers in L1 identification procedures for closely related languages.

Spoofing countermeasure for fake speech detection using brute force features

Due to the progress in deep learning technology, techniques that generate spoofed speech have significantly emerged. Such synthetic speech can be exploited for harmful purposes, like impersonation or disseminating false information. Researchers in the area investigate the useful features for spoof detection. This paper extensively investigates three problems in spoof detection in speech, namely, the imbalanced sample per class, which may negatively affect the performance of any detection models, the effect of the feature early and late fusion, and the analysis of unseen attacks on the model. Regarding the imbalanced issue, we have proposed two approaches (a Synthetic Minority Over Sampling Technique (SMOTE)-based and a Bootstrap-based model). We have used the OpenSMILE toolkit, to extract different feature sets, their results and early and late fusion of them have been investigated. The experiments are evaluated using the ASVspoof 2019 datasets which encompass synthetic, voice-conversion, and replayed speech samples. Additionally, Support Vector Machine (SVM) and Deep Neural Network (DNN) have been adopted in the classification. The outcomes from various test scenarios indicated that neither the imbalanced nature of the dataset nor a specific feature or their fusions outperformed the brute force version of the model as the best Equal Error Rate (EER) achieved by the Imbalance model is 6.67 % and 1.80 % for both Logical Access (LA) and Physical Access (PA) respectively.

Data augmentation using a 1D-CNN model with MFCC/MFMC features for speech emotion recognition

Speech emotion recognition (SER) is attractive in several domains, such as automated translation, call centres, intelligent healthcare, and human–computer interaction. Deep learning models for emotion identification need considerable labelled data, which is only sometimes available in the SER industry. A database needs enough speech samples, good features, and a better classifier to identify emotions efficiently. This study uses data augmentation to enhance the amount of input voice samples and address the data shortage issue. The database capacity increases by adding white noise to the speech signals by data augmentation. In this work, the Mel-frequency Cepstral Coefficient (MFCC) and Mel-frequency Magnitude Coefficient (MFMC) features, along with a one-dimensional convolutional neural network (1D-CNN), are used to classify speech emotions. The datasets utilized to estimate the model's enactment were AESDD, CAFE, EmoDB, IEMOCAP, and MESD. The data augmentation with the 1D-CNN (MFMC) model performed best, with an average accuracy of 99.2% for AESDD, 99.5% for CAFE, 97.5% for EmoDB, 92.4% for IEMOCAP and 96.9% for the MESD database. The proposed 1D-CNN (MFMC) with data augmentation outperforms the 1D-CNN (MFCC) without data augmentation in emotion recognition.

Characteristics of Nasal Coarticulation by Nasality Visualization System in Children with Speech Sound Disorders

Objectives: In this study, we analyzed the nasal coarticulation characteristics of children with and without speech sound disorders by comparing nasalance scores according to speech contexts. Methods: The subjects of this study were 30 children, 15 children with speech sound disorder (SSD) and 15 typically developing (TD) children, aged 4 to 6 years old. Subjects were asked to speak stimuli in different speech contexts-low vs. high vowels, nasal consonant+vowel (NV) vs. vowel+nasal consonant (VN) syllables, words vs. nonwords - and nasalance scores were measured using the Nasality Visualization System (NVS) on their speech samples. Results: First, both groups showed significantly higher naslance scores for high vowels compared to low vowels, with a significant between-group difference in high vowel. Second, both groups showed significantly higher nasalance scores for VN syllables than for NV syllables, with no between-group difference. Third, both groups showed significantly lower nasalance scores in nonwords compared to words. While there was no group difference in the words, the nasalance scores of the SSD group was lower in the nonwords, which was significantly different from that of the TD groups. Conclusion: These results suggest that the SSD group exhibits somewhat different nasal characteristics than the TD group, and furthermore, that SSD children articulate phoneme sequences more segmentally in unlearned nonword contexts.

The effect of cerebellar lesıons on language in Turkısh speakıng indıvıduals

ABSTRACT Background The cerebellum is approximately 10% of the brain volume and is responsible for attention, language, emotion and cognitive functions. Linguistic processing difficulties and cognitive impairments may be observed in individuals with cerebellar lesions depending on the location and severity of the damage. There is no study evaluating language in Turkish-speaking individuals with cerebellar lesions. Objective To determine the differences in linguistic and visuospatial skills of Turkish speaking individuals with cerebellar lesions compared to control groups, including a right hemisphere lesion group and a healthy control group. Method Fifteen patients with cerebellar lesion (without hemispatial neglect and dysarthria), 15 patients with right hemisphere lesion due to right arter cerebri media (without hemispatial neglect) and 15 healthy control were included. To clarify the language impairment specific to cerebellar lesions, we also included individuals with right hemisphere lesions, which are typically assumed not to present with aphasia. The Language Assessment Test for Aphasia (ADD), Benton Judgment of Line Orientation Test, Benton Facial Recognition Test, Single Letter Deletion Test were used. The “Accident Scene” picture was used for collecting narrative speech samples. After a three group comparison a post-hoc pairwise analysis was performed. Results Among the microstructural parameters in the narrative language analysis, the effort ratio was significantly higher in the cerebellar lesion group compared to the healthy control group. Among the non-verbal cognitive tests, the Benton Facial Recognition Test score was significantly lower in the cerebellar and right hemisphere lesion group compared to the healthy control group. Total score, naming and grammar scores in the ADD test were significantly lower in the cerebellar lesion group compared with the healthy control group. There was no difference between cerebellar and right hemisphere lesions in terms of ADD score, ADD subtests and narrative parameters. Conclusion In Turkish-speaking individuals with cerebellar lesions, verbal productivity may be interrupted by effort behaviours, and standard aphasia assessment tests may reveal impaired grammar and naming performance without a diagnosis of aphasia. In addition to reduced efficiency in verbal production, facial recognition may be impaired. We found no difference between individuals with lesions in the cerebellum and right hemisphere. This suggests that ischemic damage may affect different components of language independently of the diagnosis of aphasia.

Cross-Linguistic Nasalance Comparisons: A Review of Speech Sample Sets and Preliminary Consideration of Effect of Lexical Tone.

Nasalance is an acoustic representation of perceived nasality with proven clinical and research utility. Its validity is contingent on appropriate speech sample sets and distinct normative databases based on known impact factors such as language and phonetic environment, but little is known about the potential effects of lexical tone on nasalance. Its use in international cross-linguistic studies necessitates definition and added considerations of speech sampling protocol. Part I. A methodological review was undertaken using PRISMA (2020), the Scottish Intercollegiate Guidelines Network (SIGN) Methodology Checklist for evaluating risk of bias, and calculation of effect sizes and/or visual displays using tables showing grouping of similar data for synthesis of findings. Part II. A pilot study explored the effect of lexical tone on nasalance in Cantonese, a lexical tone language. Part I. 17 studies met the inclusion criteria. In addressing possible confounders and minimizing risk of bias, 13 studies were assigned an overall quality rating of (+) acceptable and the remaining 4, a high rating (++). For the 9-word string, there was a (non-clinical) difference of 3-5% between Swedish and Brazilian-Portuguese, and a moderate effect size for age (d=0.49); for the consonant-vowel syllables set (CV set), clinical differences across languages were identified for adults, and between 5-10% for children and adolescents. Part II. The pilot study showed a significant effect of lexical tone on nasalance where nasalance for the high-level tone 1 was significantly higher than that for high-rising tone 2, low-rising tone 5, and low-falling tone 4. There is a need for further evidence from other languages, including tonal languages, to better define the evidence and speech methodology of international cross-linguistic nasalance studies.

Overuse of familiar phrases by individuals with Williams syndrome masks differences in language processing.

We investigated whether individuals with Williams Syndrome (WS) produce language with a bias towards statistical properties of word combinations rather than grammatical rules, resulting in an overuse of holistically stored, familiar phrases. We analysed continuous speech samples from English children with WS (n = 12), typically developing (TD) controls matched on chronological age (n = 15) and TD controls matched on language age (n = 14). Alongside word count, utterance length, grammatical complexity, and morphosyntactic errors, we measured familiarity of expressions by computing collocation strength of each word combination. The WS group produced stronger collocations than both control groups. Moreover, the WS group produced fewer complex sentences, shorter utterances, and more frequent function words than chronological-age matched controls. Language in WS may appear more typical than it is because familiar, holistically processed expressions mask grammatical and other difficulties.

Putting it into words: Emotion vocabulary, emotion differentiation, and depression among adolescents.

Emotion differentiation (ED; the ability to distinguish discrete internal emotion states) may reflect or benefit from knowledge of linguistic labels. The present study uses natural language processing to examine how emotion vocabulary (EV; diversity of unique emotion terms within active vocabulary) relates to ED and depression in an adolescent sample. We tested two competing preregistered (https://osf.io/4j75w/) models regarding the EV-ED link. In the lexical facilitation hypothesis, we posited that larger EV may inform ED, perhaps resulting in larger EVs being associated with greater ED. In the emotional concision hypothesis, we theorized that ED may reflect narrower emotional experiences that are more succinctly labelled, which could result in larger EV being associated with lower ED. A community sample of adolescents (N = 241, ages 14-17, predominantly White) completed interviews, self-report measures, and ecological momentary assessments as part of a larger study conducted between 2014 and 2016. EV was derived using speech samples from transcribed recordings of life stress interviews. In line with the emotion concision hypothesis, EV and ED were inversely related for negative emotions. Moreover, larger negative EV and lower negative ED were each uniquely associated with depression, casting further doubt on whether diverse negative EVs within spontaneous language are fundamentally adaptive for emotional functioning. Replication in more diverse samples is needed to extend generalizability. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Exploring a Mobile Technology-Driven Model for Intercultural Communication Education

In the context of accelerating globalization, intercultural communication competence has become a crucial element for the success of individuals and organizations. The frequent international interactions have underscored the importance of enhancing methods for improving intercultural communication skills within the educational sector. The rapid advancement of mobile technology offers unprecedented opportunities for educational innovation. Its convenience and widespread use have made mobile device-based learning models highly attractive. Particularly, advancements in real-time speech processing technologies have provided new tools and methods for intercultural communication education. By leveraging mobile real-time speech detection and synthesis technologies, a more interactive and personalized learning experience can be achieved, thereby enhancing the efficiency and effectiveness of language learning. This study aims to explore a mobile technology-based model for intercultural communication education and is divided into three main parts: firstly, the investigation of mobile real-time speech detection technologies aimed at intercultural communication education to provide instant feedback and improvement suggestions; secondly, the exploration of mobile real-time speech synthesis technologies to generate high-quality speech samples for learners to practice against; and thirdly, the integration of the aforementioned technologies to develop a flexible, efficient, and highly interactive learning system based on mobile technology. This study is expected to not only improve the effectiveness of intercultural communication education but also provide significant references for the innovative application of educational technologies.