Phoneme Level Research Articles

Train & Constrain: Phonologically Informed Tongue Twister Generation from Topics and Paraphrases

Abstract Previous work in phonologically and phonetically grounded language generation has mainly focused on domains such as puns and poetry. In this article, we present new work on the generation of English tongue twisters - a form of language that is required to be conditioned on a phoneme level to maximize sound overlap, while maintaining semantic consistency with an input topic or phrase and still being grammatically correct. We present TwisterLister, a pipeline for generating phonologically informed tongue twisters from large language models (LLMs) that we use to generate TwistList 2.0, the largest annotated dataset of tongue twisters to date, consisting of 17K+ examples from a combination of human and LLM authors. Our generation pipeline involves the use of a phonologically constrained vocabulary alongside LLM prompting to generate novel, non-derivative tongue twister examples.We additionally present the results of automatic and human evaluation of smaller models trained on our generated dataset to demonstrate the extent to which phonologically motivated language types can be generated without explicit injection of phonological knowledge. Additionally, we introduce a phoneme-aware constrained decoding module (PACD) that can be integrated into an autoregressive language model and demonstrate that this method generates good quality tongue twisters both with and without fine-tuning the underlying language model. We also design and implement a range of automatic metrics for the task of tongue twister generation that is phonologically motivated and captures the unique essence of tongue twisters, primarily based on phonemic edit distance (PED).

Effects of Lexical Properties in L2 Chinese Compound Processing: A Multivariate Approach.

Previous studies on L2 (i.e., second language) Chinese compound processing have focused on the relative efficiency of two routes: holistic processing versus combinatorial processing. However, it is still unclear whether Chinese compounds are processed with multilevel representations among L2 learners due to the hierarchical structure of the characters. Therefore, taking a multivariate approach, the present study evaluated the relative influence and importance of different grain sizes of lexical information in an L2 Chinese two-character compound decision task. Results of supervised component generalized linear regression models with random forests analysis revealed that the orthographic, phonological and semantic information all contributed to L2 compound processing, but the L2 learners used more orthographic processing strategies and fewer phonological processing strategies compared to the native speakers. Specifically, the orthographic information was activated at the whole-word, the character and the radical levels in orthographic processing, and the phonological information at the whole-word, the syllable, and the phoneme levels all exerted contributions in phonological processing. Furthermore, the semantic information of the whole words and the constituents was accessed in semantic processing. These findings together suggest that the L2 learners are able to use cues at all levels simultaneously to process Chinese compound words, supporting a multi-route model with a hierarchical morphological structure in such processing.

Synthetic faces generated with the facial action coding system or deep neural networks improve speech-in-noise perception, but not as much as real faces.

The prevalence of synthetic talking faces in both commercial and academic environments is increasing as the technology to generate them grows more powerful and available. While it has long been known that seeing the face of the talker improves human perception of speech-in-noise, recent studies have shown that synthetic talking faces generated by deep neural networks (DNNs) are also able to improve human perception of speech-in-noise. However, in previous studies the benefit provided by DNN synthetic faces was only about half that of real human talkers. We sought to determine whether synthetic talking faces generated by an alternative method would provide a greater perceptual benefit. The facial action coding system (FACS) is a comprehensive system for measuring visually discernible facial movements. Because the action units that comprise FACS are linked to specific muscle groups, synthetic talking faces generated by FACS might have greater verisimilitude than DNN synthetic faces which do not reference an explicit model of the facial musculature. We tested the ability of human observers to identity speech-in-noise accompanied by a blank screen; the real face of the talker; and synthetic talking faces generated either by DNN or FACS. We replicated previous findings of a large benefit for seeing the face of a real talker for speech-in-noise perception and a smaller benefit for DNN synthetic faces. FACS faces also improved perception, but only to the same degree as DNN faces. Analysis at the phoneme level showed that the performance of DNN and FACS faces was particularly poor for phonemes that involve interactions between the teeth and lips, such as /f/, /v/, and /th/. Inspection of single video frames revealed that the characteristic visual features for these phonemes were weak or absent in synthetic faces. Modeling the real vs. synthetic difference showed that increasing the realism of a few phonemes could substantially increase the overall perceptual benefit of synthetic faces.

한국어교육학과 언어병리학 그리고 청각학의 학문 간 융합 연구 시론

In order to expand the research area of Korean language pedagogy, this paper attempted to explore convergence research tasks with academic fields that have the commonality of Korean, such as Speech-Language Pathology and auditory studies. Speech-Language pathology is essentially an academic field that seeks ways to solve problems for people who have communication problems. Audiology is also an academic field that focuses on how to overcome communication difficulties using speech sounds based on the perception characteristics of speech sounds. As such, the need for convergence research to solve tasks that are difficult to solve independently in each academic field is great in that they are interested in the nature of Korean and its usage patterns. In the field of Speech-Language Pathology, in order to solve problems related to language development, standards for accurately measuring the degree of language development and treating people with delayed development or disabilities are needed. Although there is a need for convergence research with the Korean Language Pedagogy on how to prepare standards and guide language functions such as reading, writing, speaking, and listening, as well as essential vocabulary and grammar dlements used in everyday life, according to age and school age, such research has not been sufficiently conducted. There is no standard for treating this because it does not clearly present clear standards for language ability development, and it is difficult to present appropriate and effective educational methods in terms of educational methodology. In this regard, it was argued that future research on the development and rehabilitation methods of language development diagnostic tools should be conducted. In the field of Audiology, research related to the development of speech perception test tools is needed. For full communication, it is necessary to develop a tool that can test the communication ability of words, phrases, sentences, and discourse units beyond the stimulation of speech sounds at the phoneme level. Although there are differences in educational methods depending on the subject of education, linguistic pathology, and audiology, the fundamental content of education is very large. In the case of researchers, there may be a tendency to think that Korean language pedagogy has conducted relatively more research related to the characteristics of the Korean language than speech-language pathology and audiology. However, it is necessary to acknowledge that the research results on the nature of the Korean language required by neighboring disciplines are insignificant, and to conduct research to identify the nature and usage patterns of the Korean language through communication and cooperation with neighboring disciplines.

Comparative study of event-related potential responses within syllables of intra and inter phoneme classes

The present study was carried out to assess the signature neural dynamics as evinced through event-related potential (ERP) responses of the P300 waveform across different inter and intra phoneme consonant classes and to quantify the differential neural dynamical response in real-time. The ERP responses to non-sense three-syllable consonantvowel transitions CVCVCV non-words were recorded on 26 healthy individuals of either sex with an age group of 18–25 years. The first syllable of the non-sense word was always fixed, and the ERP responses of the middle syllable were recorded at the phoneme level, and the ERP responses elicited by the middle syllable were evaluated. The results of the present study documented that different consonant phoneme classes (plosives, fricatives, affricatives, liquids, and nasals) elicited differential neural dynamical responses as observed through the P300 waveform response, exhibiting a significant difference in P300 waveform amplitude across the fricatives vs. nasals experimental design.

Shared Loanword Recognition in German-English Bilinguals: The Role of Metrical Phonology.

The role of phonology in bilingual word recognition has focused on a phonemic level especially in the recognition of cognates. In this study, we examined differences in metrical structure to test whether L1 metrical structure influences the processing of L2 words. For that, we used words of Romance origin (e.g., reptile, signal), which both German and English have borrowed extensively. However, the existing metrical patterns are not identical nor are the borrowed vocabularies the same. Rather, those identical words differ systematically in their foot structure. We conducted a cross-modal form fragment priming EEG experiment (auditory-visual) with German native speakers who were highly proficient in English. Both behavioral and ERP results showed an effect of the native phonology and the loan status, that is, whether the loan exists only in the speaker's L2 or is shared across languages. Priming effects (RTs) were largest for nonshared loanwords indicating some interference from German (L1). This was also evident in a reduced N400 but only if the metrical structure aligned with German patterns for Germanic words, that is, two light syllables as in pigeon. If the words exist in both languages, metrical structure also mattered shown by the modulation of different ERP components across conditions. Overall, our study indicates that metrical phonology plays a role in loanword processing. Our data show that the more similar a word is in terms of its metrical phonology across L1 and L2, the more effortful the processing of a word within a priming paradigm indicating interference from the L1 phonology.

W2VC: WavLM representation based one-shot voice conversion with gradient reversal distillation and CTC supervision

Non-parallel data voice conversion (VC) has achieved considerable breakthroughs due to self-supervised pre-trained representation (SSPR) being used in recent years. Features extracted by the pre-trained model are expected to contain more content information. However, in common VC with SSPR, there is no special implementation to remove speaker information in the content representation extraction by SSPR, which prevents further purification of the speaker information from SSPR representation. Moreover, in conventional VC, Mel-spectrogram is often selected as the reconstructed acoustic feature, which is not consistent with the input of the content encoder and results in some information lost. Motivated by the above, we proposed W2VC to settle the issues. W2VC consists of three parts: (1) We reconstruct feature from WavLM representation (WLMR) that is more consistent with the input of content encoder; (2) Connectionist temporal classification (CTC) is used to align content representation and text context from phoneme level, content encoder plus gradient reversal layer (GRL) based speaker classifier are used to remove speaker information in the content representation extraction; (3) WLMR-based HiFi-GAN is trained to convert WLMR to waveform speech. VC experimental results show that GRL can purify well the content information of the self-supervised model. The GRL purification and CTC supervision on the content encoder are complementary in improving the VC performance. Moreover, the synthesized speech using the WLMR retrained vocoder achieves better results in both subjective and objective evaluation. The proposed method is evaluated on the VCTK and CMU databases. It is shown the method achieves 8.901 in objective MCD, 4.45 in speech naturalness, and 3.62 in speaker similarity of subjective MOS score, which is superior to the baseline.

Phonemic structure of the root morpheme in the Gothic language (a complex quantitative analysis of lexicographic sources)

The article deals with the problems of the phonemic structure of the root morpheme in Gothic. Using a systematic sampling procedure, the words from a Gothic dictionary were selected to form the register with 382 root morphemes. The study employed a quantitative research with statistical data analysis toinvestigate such characteristics of the Gothic root morpheme as its length in phonemes, distribution, combinability and phonotactics of vowel and consonant phonemes within CV root patterns, their constructive potential and symmetry. The investigation has revealed several fundamental regularities that characterize the Gothic root morpheme organizing on the phonemic level. The results of the study have shown that the Gothic language has roots ranging in length from two to nine phonemes, but four-fifths of the analyzed roots consisted of 3-4 phonemes, the average length of the Gothic root in phonemes is 3.55, the multi-phoneme roots are rare and represented mainly by lexical items of non-Germanic origin. The analysis of the realized canonical forms has shown that the Gothic language system uses only 2% of the theoretically possible forms for the construction of root morphemes at the phonemic level. The vast majority of these roots function as dependent roots, only 5% of them function as independent roots that form complete lexical units. The canonical forms of roots in the Gothic language have different modeling power; the most productive among them are three structural types (CVC, CVCC, CCVC), which form almost four fifths of all roots in the studied array. The phonemic structure of the root morpheme is characterized by a significant consonantal saturation: the initial and final components of most canonical forms are formed by consonants, and the medial ones by vowels. Structures of this type describe almost 90% of the total number of roots in the study sample. It has been proven that the phonemic structure of Gothic roots is characterized by symmetry, with the symmetrically constructed roots being mostly characterized by the mirror type of symmetry.

Automatic Detection System for Velopharyngeal Insufficiency Based on Acoustic Signals from Nasal and Oral Channels.

Velopharyngeal insufficiency (VPI) is a type of pharyngeal function dysfunction that causes speech impairment and swallowing disorder. Speech therapists play a key role on the diagnosis and treatment of speech disorders. However, there is a worldwide shortage of experienced speech therapists. Artificial intelligence-based computer-aided diagnosing technology could be a solution for this. This paper proposes an automatic system for VPI detection at the subject level. It is a non-invasive and convenient approach for VPI diagnosis. Based on the principle of impaired articulation of VPI patients, nasal- and oral-channel acoustic signals are collected as raw data. The system integrates the symptom discriminant results at the phoneme level. For consonants, relative prominent frequency description and relative frequency distribution features are proposed to discriminate nasal air emission caused by VPI. For hypernasality-sensitive vowels, a cross-attention residual Siamese network (CARS-Net) is proposed to perform automatic VPI/non-VPI classification at the phoneme level. CARS-Net embeds a cross-attention module between the two branches to improve the VPI/non-VPI classification model for vowels. We validate the proposed system on a self-built dataset, and the accuracy reaches 98.52%. This provides possibilities for implementing automatic VPI diagnosis.

Outline of the problem of developing the phonetic level of speech in older preschool children with logopathology

The scientific article presents the theoretical position of modern research on the content of the phonetic level of speech. Attention has been focused on the problem of studying the components of the phonetic level of speech in older preschool children with normotypical psychophysical development and speech disorders. The phonetic level of speech in children with logopathology was experimentally investigated and a comparative analysis was carried out with peers with normotypical psychophysical development. The purpose of the study is to study the current state of formation of the phonetic level of speech in older preschool children with speech pathology. The task of the research: scientific and theoretical substantiation of scientific positions regarding the development of the phonemic level of speech in children with normotypical psychophysical development and outlining the features of its development in children with logopathology; definition of tasks and evaluation criteria for studying the phonemic level of speech; carrying out a comparative analysis of the results of the study of the state of formation on the development of the phonemic level of speech in children of older preschool age with normotypical psychophysical development and with logopathology. The methods of research are tasks aimed at researching the components of the phonetic level of speech, namely phonetic perception (the ability to differentiate phonemes that are similar in sound, distinguish a sound from a group of sounds, determine the sound in a word), phonetic analysis (the ability to determine the first and last sounds in a word, as well as the number of sounds in a word) and phonetic representation (the ability to correlate sounds with the name of an object) in older preschool children with normotypical psychophysical development and speech disorders. The results of the ascertainment experiment indicate that there are significant differences in the formation of the phonemic level of speech between the groups of the studied older preschool children with logopathology and those with normotypical psychophysical development. Phonemic processes in preschoolers with logopathology, as compared with normotypical indicators, develop with a delay, which leads to a number of errors: during phonemic perception, children do not recognize close-sounding sounds; do not know how to distinguish a sound from a group of sounds or in a word; during phonetic analysis – they incorrectly detect the first and last sounds in a word (name the first syllable or part of a word), make a mistake in determining the number of sounds in a word; during phonemic representation - cannot find the sound in the name of the object.

Towards a Vocal and Acoustic Description of Kapa Haka

ObjectiveThe purpose of this study is to analyze the voice used in kapa haka, a contemporary indigenous vocal performance from New Zealand, which includes the well-known haka. This is the first study of its kind and is a preliminary investigation into the vocal and acoustic description of kapa haka. A significant goal of this study is to contribute ideas and potential definitions of vocal qualities to the community of kapa haka trainers that were specific to the genre. This strengths-based project raises up these vocal practices as legitimate and authentic colors within a vocal tradition that has seen generational learning disrupted by colonial interventions and is now flourishing within the community. MethodsEight kapa haka performers (three females, five males) were involved in the study, they were all experienced performers; furthermore, two had formal classical voice training. They were individually recorded speaking and performing three different genres of kapa haka (mōteatea, waiata, and haka); all recordings were in te reo Māori (the Māori language). In addition, electroglottograph (EGG) signals were collected. An auditory-perceptual evaluation of the kapa haka voice was completed by three singer–researcher–pedagogues familiar with Western and non-western genres of singing. They all have experience appropriately collecting and analyzing data from indigenous communities, and they all understand the sociopolitical context of the vocal genre within the local colonial history. A specific evaluation instrument was created, and the results were validated. The acoustic and time-aligned EGG data was annotated at the phoneme level, and the signal analysis was performed in MATLAB. Averaged EGG pulses from /a/ segments were investigated, along with long-term average spectrums of the performances obtained from both the audio signal and the EGG signals. ResultsThe perceptual analysis suggests the biggest difference in vocal styles was between the haka and the other two genres (and speech). The acoustic and EGG results support these findings. ConclusionsCommon characteristics, perceptually and acoustically, were identified in the kapa haka performance styles across the eight performers.

Perceiving speech during orthographic syllable recognition: Beyond phonemic identity

In the cue-distractor paradigm, individuals observe a spoken distractor syllable while responding to a visual cue referring to a syllable. When the task is to utter the cued syllable, distractors sharing fewer subphonemic properties with the cued syllable (below the level of phonemes) lead to slower reaction times (RTs), indicating representations involved in speech perception and production are closely linked. The present study investigated whether a subphonemic level of representation is involved when the task was to manually indicate (but not produce) an orthographically cued syllable. Results revealed RT modulations closely mirroring those reported previously for uttered responses. In an additional experiment, phonetic variants of phonologically identical distractors were presented, but RT modulations were unaffected by this manipulation. The present findings indicate that perceiving speech accesses a relatively detailed phonological level of representation which is closely aligned with representations pertinent in orthographic syllable recognition and in speech production.

Hybrid Approach to Detect Prolonged Speech Segments

In the last 10 decades various methods have been introduced to detect prolonged speech segments automatically for stuttered speech signals. However less attention has been paid by researches in the detection of prolongation disorder at the parametric level. The aim of this study is to propose a hybrid approach to detect the prolonged speech segments by combining various spectral parameters with their recognition accuracies for the reconstructed speech signal. The paper presents prolonged segments detection by considering the parameters individually, combining various spectral parameters, validation of prolongation detection system, MFCC feature extraction process, basic model accuracies for the reconstructed signals. The proposed methods are simulated and experimented on UCLASS derived dataset. Obtained results are compared with the existing works of prolongation detection at parametric and word level. It is observed that hybrid parameters yield 92% of recognition rate for larger frame sizes of 200ms when modeled with SVM. The results are also tabulated and discussed for various metrics like sensitivity, specificity and accuracy metrics in detecting the prolonged segments. The study also focuses on the prolongation characteristics of vocalized and non-vocalized sounds at phoneme level. The detection accuracy of 71% is observed for Vocalized prolonged vowel phonemes over non-vocalized prolonged signal. Objectives: The objective of this work is to propose a hybrid algorithm to detect prolonged segments automatically for speech signal with prolongation disorder. The other objective is to evaluate the obtained spectral parameters performances by applying to various evaluation metrics and models to compute the recognition accuracy of a reconstructed signal. The objective is further extended to bring out the importance of variable frame size concept and to analyze the variations in vocalized and non-vocalized sounds. Methods: The methods adopted to detect prolonged speech segments are discussed at two levels namely at the preprocessing and modeling levels. The Preprocessing level is discussed by applying various parameters at an individual level, hybrid level by combing the Centroid, Entropy, Energy, ZCR parameters and MFCC feature extraction method. A new method has been applied using Specificity, Sensitivity and accuracy metrics to validate the prolongation detection model performance. In modeling level, the above parameters are discussed by applying evaluation metrics for the clustering and classification models like K-means, FCM and SVM. The performance of these methods is considered for evaluating and estimating the prolonged segment detection accuracy of the reconstructed speech signals of vocalized and non-vocalized sounds. All these methods are discussed in detail in the following sections. Findings: Hybridizing the spectral parameters to detect the prolonged speech segment automatically is a major finding of this work. It is also found that Specificity, sensitivity and accuracy metrics plays a major role in designing and validating the prolongation detection model. From the further experiments it is identified that the hybrid and verification metrics suits better for vocalized and non-vocalized sounds when larger frame lengths are considered. SVM has been found to perform better for all the above considerations. Novelty: As per Literature survey it is observed that individual and few parameters are applied to detect the prolongation. But works are not addressed on applying or combining more than two parameters to detect the prolonged speech segments. The novelty of this work lies in selecting and combining the spectral parameters at the preprocessing stage to detect the prolongation disorder. Spectral centroid and entropy are considered as appropriate parameters along with ZCR and Energy parameters. Hence hybridizing these parameters results in a novelty to propose an automatic prolongation detection system. Novelty is further brought by applying Specificity, sensitivity and accuracy metrics to build and evaluate the detection system for vocalized and non-vocalized prolonged sounds.

Phonological Awareness and Word Reading Fluency Among Young Saudi Learners of English

The current study explored the relationship between phonological awareness and word reading fluency among young Saudi English as a Foreign Language (EFL)learners. Two tests were completed by thirty young Saudi EFL learners. The first test comprised 25 questions related to phonological awareness. In the second test, the learners were asked to read a list of English words in one minute, and each student was given a score based on how many words they read per minute. The students scored relatively low on both tests, indicatinga deficiency in phonological awareness and difficulty in word reading. The scores in both tests were significantly correlated, however, a regression analysis showed that only phonological awareness at the phonemic level was a significant predictor of the students’ performance in the word reading fluency test. The results clearly indicate that Saudi EFL learners could benefit from phonemic awareness training, and future studies should empirically examine its potentialfor improving Saudi EFL learners’ reading fluency.

Securing Liveness Detection for Voice Authentication via Pop Noises

Voice authentication has been increasingly adopted for sensitive operations on mobile devices. While voice biometrics can distinguish individuals by their spectral features (such as voiceprints), they are known to be prone to spoofing attacks, where malicious attackers can use pre-recorded or synthesized samples from legitimate users or impersonate the speaking style of the targeted user to deceive the voice authentication system. In this paper, we design and implement a novel software-only anti-spoofing system on smartphones. Our system leverages the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">pop noise</i> , which is generated by the user’s oral airflow when speaking the passphrase close to the microphone. The pop noise is delicate and subject to user diversity, making it hard to be recorded by replay attacks beyond a certain distance or to be imitated precisely by impersonators. Specifically, we design a new pop noise detection scheme to pinpoint pop noises at the phonemic level, based on which we establish a theoretical model to calculate the sound pressure level from the speech signal in order to get the estimated pressure signal, and then analyze the consistency with the actual pressure signal extracted from the pop noise. Furthermore, we calculate the similarity score of the unique sequences which describe the individually unique relationship between pop noises and phonemes to resist spoofing attacks. Our evaluation on a dataset of 30 participants and three smartphones shows that our system achieves over 94.79% accuracy. Our system requires no additional hardware and is robust to various factors including authentication angle, authentication distance, the length of passphrase, ambient noise, etc.

Extracting Spatial Muscle Activation Patterns in Facial and Neck Muscles for Silent Speech Recognition Using High-Density sEMG

Human is capable of producing versatile and precise speech owing to the complex neuromuscular systems that control the movement of facial and neck muscle groups. Silent speech recognition based on surface electromyography (sEMG) has been widely researched, which enables speech communication in the absence of an audible acoustic signal. However, the differences in the activities of facial and neck muscle groups under different phoneme pronunciation conditions have not been well investigated. The present work aims to quantify the macroscopic spatial patterns of the muscle activations at the phoneme level during the movement of facial and neck speech articulator muscle groups (SAMGs) using 320-channel high-density (HD) electrode arrays. 10 subjects performed 14 vowel speech tasks and 15 consonant speech tasks in audible speech and silent speech modes, respectively. Specifically, the root-mean-square heat maps of sEMG signals were used to characterize the SAMGs activation patterns. The results show that the muscle activation regions are overlapped under different phoneme pronunciation conditions and symmetric between the left and right sides of face and neck with the average correlation coefficient > 0.9. In addition, the best result can achieve a classification accuracy of 85.78% for consonant and 79.42% for vowel in silent speech mode. The overall results can provide a reference to the hardware or electrode design of portable device for practical use in future work.

A Stylistic Study in Surat Al-Baynah

The study uses a descriptive-analytical approach to provide a close reading of the stylistics of Surat Al-Bayyinah. The researchers begin by defining the concept of stylistics and levels of analysis, in addition to situating Surat Al-Bayyinah in its historical context. Phonologically, they track the role of sounds and alliteration in establishing rhythm within the Surat’s verses. Structurally, they analyze the semantic relations that link the linguistic elements of the Surat and track the rhetorical connotations of its verses. Among the findings of the study is the predominance of the high voice over the bass voice at the phonemic level, and the frequency of verbal sentences compared to nominal sentences. The semiotic level perfectly resonated with the phonological and structural dimensions of the Surat.

Development of Small Vocabulary Continuous Speech-to-Text System for Kannada Language/Dialects

Objectives: To develop a speech-to-text (STT) system using Kaldi speech recognition toolkit for continuous Kannada language/dialects. Methods: A continuous Kannada speech data is collected from 100 speakers/farmers of Karnataka state in field. The lexicon/dictionary and set of phonemes for Kannada language/dialects are created and transcribed the collected speech data using transcriber tool. The ASR models are developed at different phoneme levels using Kaldi. Findings: In this work, an effort is made to develop a robust small vocabulary STT system for continuous Kannada language using Kaldi. The various acoustic modelling techniques are used to develop a robust ASR model and achieved a word error rate (WER) of 0.23%. The performance of the developed ASR model is compared with existing works and analyzed by offline speech recognition. Novelty: Many STT systems have been developed for Indian and International languages/dialects, but not for Kannada language. This work is first of its kind using Kaldi in Kannada language under the constraints of limited data. The developed ASR model could be used further in the development of end-to-end ASR system for speech processing applications. Keywords: Automatic Speech Recognition (ASR); Word Error Rate (WER); Continuous Kannada Speech Data; Kannada Language/Dialects; Lexicon

Decoding silent speech from high-density surface electromyographic data using transformer

Recent silent speech recognition (SSR) studies based on surface electromyography (sEMG) have been conducted by classifying a finite number of words or phrases without sufficient understanding of temporally semantic information compared to sequential decoding at a fine-grained syllable or phoneme level. This paper presents a syllable-level sequential decoding method using a transformer model for sEMG-based SSR. The proposed method consists of a transformer model and a language model. The input sEMG data was first translated into a sequence of syllable-level decisions by the transformer model. Then, these sequential syllable-level decisions were tuned as a final syllable sequence to approximate natural language through the language model. To verify the effectiveness of the proposed method, experiment data were recorded using two high-density electrode arrays with 64 channels from a total of eight subjects during subvocally reading a corpus of 33 Chinese phrases generated from a dictionary of 82 syllables. The proposed method achieved the lowest character error rate of 5.14 ± 3.28 % and the highest phrase recognition accuracy of 96.37 ± 2.06 %, and it significantly outperformed other common methods for sEMG-based SSR. These findings demonstrated the feasibility and usability of the proposed method for practical SSR applications.

English Speech Scoring System Based on Computer Neural Network

In English phonetics teaching, in order to improve students' English phonetics quality, a computer neural network based English phonetics scoring method is proposed. First, the frequency domain spectrogram is used as the data input to construct a convolutional neural network model at the word and phoneme levels to detect speech similarity. Then the original sound time domain waveform is used as the data input, which is converted into text through neural network to detect the text difference. Finally, we combine the two with the assigned weight to give a relatively objective comprehensive pronunciation score. The simulation results show that the method is accurate and practical, and can promote the standardization of students' English pronunciation.