Accurate Speech Research Articles

Emotion Recognition from Spontaneous Tunisian Dialect Speech

Emotional expressions are a fundamental aspect of human communication, with speech being one of the most natural modes of interaction. Speech Emotion Recognition (SER) is a significant research topic in Natural Language Processing (NLP), aimed at identifying emotions such as satisfaction, frustration, and anger from speech audio using multiple classifiers. This paper presents a method to emotion recognition from spontaneous Tunisian Dialect (TD) speech, marking the first work in the SER field to utilize spontaneous speech for emotion recognition in this dialect. The dataset was created from freely available YouTube videos across multiple domains and labeled with four perceived emotions: anger, satisfaction, frustration, and neutral. To address the data scarcity issue, we implemented data augmentation techniques, specifically Vocal Tract Length Perturbation (VTLP). The preprocessing of the speech signals involved cleaning the data from ambient and unwanted noises. We extracted and selected various spectral features, including mel-frequency cepstral coefficients (MFCC) and Linear Prediction Cepstral Coefficients (LPCC). Subsequently, we applied several classification methods: Support Vector Machine (SVM), Bidirectional Long Short-Term Memory (BiLSTM), Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Random Forest. Our experiments demonstrated that the Random Forest classifier achieved the highest F-score of 58.75%. The results were thoroughly discussed, analyzed, and compared across the five models using different feature extractions. This study provides valuable insights and advancements in the SER field, particularly for the TD, future research directions for improving emotion recognition systems.

The Karelian language landscape in the dialectometric paradigm

Karelian dialectal speech has been studied by linguists of Finnic languages in Russia and elsewhere for a century and a half, but such issues in Karelian dialectology as the fuzzy distribution of dialectal units over the territory, selection of the fundamental principle for dialectal division, determination of the language status of specific varieties, etc., have so far remained unresolved. This article summarizes the results of a study of Karelian dialectal speech based on archival materials from the 1930s–1970s in the “Dialectological Atlas of Karelian” (1997) performed using the dialectometric method of cluster analysis: the main contrastive phenomena of phonological, morphonological, morphological and lexical linguistic levels are listed, the distribution ranges are described and a typology of their members is provided, the current three-level dialect classification and the periodization of the history of the Karelian language are presented. Three supra-dialects are outlined on the dialect map: Karelian Proper, Livvi, and Ludic. They are distinguished on the basis of isoglosses of the morphological and morphonological systems inherited from the early period in the development of the language. The Karelian Proper supra-dialect rests upon Balto-Finnic Protolanguage and pan-Karelian novel traits; Livvi shows Balto-Finnic archaisms, features coming from the Old Veps substrate, which has affected different language levels, Old Karelian innovations, and its own dialectal features; Ludic incorporates Old Karelian innovations and a pronounced Old Veps substrate. The Karelian Proper supra-dialect was subdivided into three dialects: northern, southern (variants spoken in Central Russia), and a transitional dialect occupying an intermediate position between supra-dialects. Ludic is differentiated into two dialects: the original Ludic and the Pryazha dialect (Ludic variants spoken in the Pryazha District of Karelia), which was heavily influenced in later periods by neighboring Livvi sub-dialects. Variants of the Livvi supra-dialect proved to be relatively uniform. The making of the dialects was shaped by late Old Karelian dialectal differences and intensive contacts in the border areas of the supra-dialects, whereas the twenty sub-dialectal groups became differentiated through relatively recent phonetic and lexical innovations and contact influence of the neighboring languages.

The effect of subtitle format on cognitive load and comprehension: an eye-tracking study

ABSTRACT In most universities in South Africa, English is still preferred as the medium of instruction. However, most schools in South Africa have adopted a home-language education approach. Those school students then have a lower English proficiency compared to other students from English schools. Providing these students with instructional assistance with subtitles is essential to their learning. In recent years, advances in automatic speech recognition systems made it possible to automatically generate subtitles. This automation does, however, come with its own issues, namely: speech inaccuracy, timing issues of subtitles, etc. This study will investigate the effect of automatically generated subtitles on comprehension and cognitive load. Findings indicated that the subtitle types differed in the presentation speed of consecutive subtitles. This significantly influenced the comprehension of the participants that had automatic subtitles by 10%. Professionally creating subtitles is expensive and time-consuming, but also has more advantages and positive effects than automatic subtitles.

Language awareness survey among university students

In this study we examined the linguistic awareness and dialect attitudes of university students studying at the University of Nyíregyháza. It is particularly important that the future teachers have a positive attitude towards the varieties of the Hungarian language and share their knowledge with their learners in an effective way. However, the results of various research studies show that in many cases, mother tongue education in schools does not build on a multi-standard approach to language varieties, and as a consequence, non-standard varieties are perceived more negatively than standard ones. The aim of this research is to find out what the university students in the research study know about the regional variation of language, conscious language use and how they view dialect speech. The study was conducted through an online questionnaire survey involving 70 respondents. As a starting point for the research, it was assumed that the respondents’ dialect attitudes and language awareness would be determined by their experiences at school, and thus negative attitudes towards dialect speech would be observed in their value judgments. It was also assumed that respondents would be less accepting of the occurrence of dialect phenomena in their own language use and in the speech habits of others in their locality. The results showed that the majority of the students surveyed did not have a negative opinion of dialect phenomena, so the first hypothesis was not confirmed. However, the second hypothesis was partially confirmed, as far fewer respondents acknowledged the occurrence of dialect phenomena in their own language use than in the speech of the people in their locality. Overall, the majority of respondents to the survey have a positive attitude towards linguistic diversity and strive to use the language consciously. It is hoped that the students surveyed here will adopt the right approach to teaching about linguistic diversity in their teaching practice, thus positively influencing learners’ attitudes towards dialects and linguistic diversity.

Is L2 pronunciation affected by increased task complexity in pronunciation-unfocused speaking tasks?

Abstract This study examines the effects of task complexity on second language (L2) pronunciation accuracy and global pronunciation measures in pronunciation-unfocused tasks and assesses the relationship between acoustic and listener-based pronunciation measures. Eighty-two Catalan/Spanish learners of English performed simple and complex versions of a problem-solving monologic speaking task, for which the oral stops /p, t, k/ and vowel contrasts /iː/-/ɪ/ and /æ/-/ᴧ/ were embedded in the lexical items used to perform the task. Pronunciation accuracy was gauged through acoustic measurements of laryngeal timing (voice onset time), vowel contrastiveness and nativelikeness (Mahalanobis distances), and native speakers’ ratings of comprehensibility and accentedness. Results revealed detrimental effects of increased task complexity on the productions of oral stops and speech comprehensibility and accentedness; however, no consistent task complexity effects were found on vowel accuracy. The analysis also revealed an association between segmental accuracy and global dimensions of L2 speech.

Emotion Recognition in Kurdish Speech from the Sorani Dialect Corpus

Given the increasing need for interactive human-computer applications, the field of employing machine learning algorithms to discern emotions from speech has seen a substantial surge in interest. While emotion recognition systems have made substantial progress in languages like German, English, Spanish, Dutch, and Danish, the availability of comprehensive datasets for the Kurdish language remains notably limited. This paper addresses this gap by focusing on emotion recognition in Sorani Kurdish dialect speech data, which was carefully gathered from openly available videos from the YouTube platform and categorized into four clear supposed emotions: neutral, sadness, happiness, and anger. The study applied both natural Mel Spectrogram and Mel-Frequency Cepstral Coefficient (MFCC) features for various spectrals, followed by the classification models K-Nearest Neighbor (KNN), Multi-Layer Perceptron (MLP), and Support Vector Machine (SVM) to evaluate the results. By closely examining and contrasting the results of using several methods for feature extraction, it was found that SVM obtained a higher accuracy, reaching as much as 85.57%. This is so much more than the first Kurdish emotion classification technique for the recognition of the emotion of the words.

L1 listeners’ evaluations of LX speech: the role of listener expectations and personality traits

ABSTRACT LX speech tends to receive less favourable judgments, which can be attributed to its non-standard linguistic properties (e.g. foreign accents, grammatical errors). However, the evaluations of LX speech could also be affected by listeners’ expectations and their individual differences. This study extends research on the role of listeners’ expectations (cued by speech accent) and personality traits (Big Five), by examining the degree to which and how accent, grammaticality and listener personality interact to determine L1 listeners’ evaluations of LX speech. Sixty L1 English speakers (30 female) rated the acceptability of 40 English speech stimuli in a British or Polish accent; half of the stimuli were filled with English article errors. The listeners’ personality traits were self-reported via a short questionnaire of the Big Five factors. The significant interaction of accent, grammaticality and listener personality suggests that the occurrence of grammatical errors is less penalised in foreign-accented speech (vs. native-accented speech), unless the L1 listeners are highly conscientious or highly extraverted. The theoretical (expectation mismatch effect, Interpersonal Circumplex model), methodological (considering non-linear models and the moderation of contextual factors for personality effects) and potential practical implications (awareness training of personality effects) of the results are discussed.

Peningkatan Keterampilan Bercerita Melalui Kartu Gambar pada Mahasiswa Program Studi Pendidikan Anak Usia Dini di IAIS Sambas

Storytelling is one of the habits of ancient people in conveying a message or telling an activity that has happened. A person's storytelling skills are closely related to the speaking skills that the person has. There are two important elements that must be mastered by students in storytelling, namely linguistics and the elements of what is told. Accuracy of speech, grammar, vocabulary, fluency and smoothness, illustrate that students have good speaking skills. Departing from the phenomena in the field, researchers see that students' storytelling skills are still lacking. This study aims to explore the improvement of storytelling skills in students of early childhood education study programs using picture cards. The method used is a quantitative approach with a descriptive type of research. The results show a significant increase in storytelling skills using picture cards.

Gender and age biases in the assessment of speech accuracy: A study of speech-language clinicians' ratings of /s/ accuracy.

Typically developing children assigned male at birth (AMAB) and children assigned female at birth (AFAB) produce the fricative /s/ differently: AFAB children produce /s/ with a higher spectral peak frequency. This study examined whether implicit knowledge of these differences affects speech-language pathologists'/speech and language therapists' (SLPs'/SLTs') ratings of /s/ accuracy, by comparing ratings made in conditions where SLPs/SLTs were blind to children's sex assigned at birth (SAB) to conditions in which they were told this information. SLPs (n = 95) varying in clinical experience rated the accuracy of word-initial /s/ productions (n = 87) of eight children with speech sound disorder in one of four conditions: one in which no information about the children was revealed, one in which children's SAB was revealed, one in which children's age was revealed, and one in which both were revealed. Despite there being no statistically significant differences between AFAB and AMAB children's /s/ production in researcher-determined accuracy or in one acoustic characteristic, spectral centroid, SLPs in all four conditions judged the /s/ productions of AFAB children as more accurate than AMAB children. Listeners were significantly less likely to judge the productions of AMAB children to be inaccurate in the conditions in which age or age and SAB were revealed. These effects were consistent across SLPs with greatly varying levels of clinical experience. Knowing or imputing children's age and SAB can affect ratings of /s/ accuracy. Clinicians should be mindful of these potential effects. Future research should understand how expectations about sociolinguistic variation in speech affect appraisals of their speech and language. What is already known on the subject Adult men and women produce /s/ differently. A consensus is that these differences reflect sociolinguistic gender marking, rather than being the passive consequence of vocal-tract differences. Recent studies have shown that children assigned female at birth (AFAB) and those assigned male at birth (AMAB) produce /s/ differently in ways that mirror the differences between adult men and women, and which presumably reflect gender marking. What this paper adds to existing knowledge We asked whether US-based speech-language pathologists' (SLPs) ratings of the accuracy of /s/ differ depending on whether they are rating an AFAB or an AMAB child, and whether these differences are greater in conditions in which people are told the sex assigned at birth of the child being rated. We found that SLPs were more likely to judge AFAB children's /s/ productions to be more accurate than AMAB children's, even though the productions from the AMAB and AFAB children that were used as stimuli were matched for accuracy as determined by trained researchers. What are the clinical implications of this work? SLPs/speech-language therapists should be sensitive to the influence of social variables when assessing /s/. SLPs/speech-language therapists might rate children's productions differently depending on whether they believe they are rating an AFAB or an AMAB child.

Smartphone automated motor and speech analysis for early detection of Alzheimer's disease and Parkinson's disease: Validation of TapTalk across 20 different devices.

Smartphones are proving useful in assessing movement and speech function in Alzheimer's disease and other neurodegenerative conditions. Valid outcomes across different smartphones are needed before population-level tests are deployed. This study introduces the TapTalk protocol, a novel app designed to capture hand and speech function and validate it in smartphones against gold-standard measures. Twenty different smartphones collected video data from motor tests and audio data from speech tests. Features were extracted using Google Mediapipe (movement) and Python audio analysis packages (speech). Electromagnetic sensors (60Hz) and a microphone acquired simultaneous movement and voice data, respectively. TapTalk video and audio outcomes were comparable to gold-standard data: 90.3% of video, and 98.3% of audio, data recorded tapping/speech frequencies within±1Hz of the gold-standard measures. Validation of TapTalk across a range of devices is an important step in the development of smartphone-based telemedicine and was achieved in this study. TapTalk evaluates hand motor and speechfunctions across a wide range of smartphones.Data showed 90.3% motor and 98.3% speech accuracy within +/-1Hz of gold standards.Validation advances smartphone-based telemedicine for neurodegenerative diseases.

Speech production skills of bilingual children using cochlear implants

Abstract Evidence is lacking on the impact of bilingualism on the speech skills of children with cochlear implants (CIs). This study described the speech production of children with CIs acquiring French and one or more additional spoken languages. Four groups of children aged 4–11 were included: bilinguals (n = 15) and monolinguals (n = 14) with CIs and bilinguals (n = 14) and monolinguals (n = 20) with typical hearing. Data were collected about the percentage of consonant correct (PCC) and vowel correct (PVC) produced in French and intelligibility in all languages they spoke. Bilingual and monolingual children with CIs had comparable speech accuracy in French, but the pattern differed, impacting PCC for bilinguals and PVC for monolinguals. Most children with CIs had accurate and intelligible speech in French, but few bilingual children with CIs were highly intelligible in their home language. Therefore, bilingualism did not impede the speech production outcomes of bilingual children with CIs in the language of the wider community.

Augmented dialectal speech recognition for AI-based neuropsychological scale assessment in Alzheimer’s disease

Alzheimer’s disease (AD) is a prevalent and widespread neurodegenerative disorder among the older adult population worldwide. Among the numerous cognitive screening methods, neuropsychological scale assessments (NSA) are the widely utilized screening tools in clinical practice. The NSA places significant emphasis on speech-related questions, and thus the role of Automatic Speech Recognition (ASR) technology in AI-based NSA evaluations becomes particularly critical. However, the majority of ASR research pays limited attention to the study of speech recognition for various dialects. One of the primary reasons and main challenges for the scarcity of research in dialectal speech recognition is the limited availability of annotated dialectal data. Furthermore, the older adults’ unclear pronunciation leads to erroneous recognition of similar phonetic words by ASR. To overcome the challenges of limited annotated dialect data in various regions and the issues related to confusable pronunciation of older adults, we propose a dialectal collaborative encoder mechanism and a machine reading comprehension augmented model (MRCAM) that automatically corrects the texts in the NSA. The dialectal collaborative encoder mechanism utilizes layer frozenness to transfer recognition ability to the dialectal model. Meanwhile, the MRCAM model fills the gap by combining speech recognition algorithms with natural language processing (NLP) technology to recognize the dialects spoken by older adults. The experimental evaluation results indicate that the proposed augmented model achieves over 90% consistency with the physician’s judgment, providing the premise of the proposed methods for the task of automated screening for AD based on ASR.

Матеріали до словника традиційних народних промислів (гончарська лексика подільських говірок)

The article analyzes lexical material for the Dictionary of Traditional Folk Crafts. The proposed materials reflect the peculiarities and regional differences of the pottery terminology of Podillia. The relevance of the article is due to the fact that many problems of dialect lexicology have not yet been solved, in particular, a national dictionary of dialects and a dictionary of individual regions of Ukrainian dialects have not been created. The nomination of phenomena and concepts of traditional folk culture is poly-informative, and their fixation and preservation remain an urgent task of modern dialectology. The sources of the factual material are recordings of live folk speech made by the author in 12 (twelve) settlements of Podillia in 1998-1999. The purpose of the research is to review the history of Podillia pottery as a folk craft, as well as to present a fragment of a dictionary of pottery terminology, and to introduce a name to denote the realities of pottery into scientific circulation. The article uses general scientific methods of research: analysis, synthesis and observation, which made it possible to identify and systematize colloquial material; the descriptive method was used in the process of analyzing the vernacular speech. Special linguistic methods of research: questionnaires and interviews contributed to the recording of dialect speech, contextual and interpretive analysis allowed to identify transformations of the semantic scope of lexemes functioning in the live speech of speakers of Podillia dialects, linguo-geographical method made it possible to illustrate the habitat of the recorded names. The result of the research is the introduction of lexical material collected during the expedition into scientific circulation. The work on the dictionary of pottery terminology is planned to be continued, streamlining the previously collected vocabulary. Originality. Studying the cultural characteristics of Ukrainians allows not only expanding general knowledge, but also passing on the nation's invaluable experience to future generations. The vocabulary of traditional crafts is a unique basis for enriching the language, a part of cultural and social memory. It is also important to compile dictionaries, as the recording of traditional folk culture concepts is an urgent issue not only for ethnology but also for linguistics. Conclusion. The pottery vocabulary of Podillia dialects is a complex multifaceted system with a clear structure. The study of this vocabulary proves that in terms of area, it shows tendencies towards spatial opposition: some of the lexemes are not found in neighboring dialects (brashne, bohonok, vedrovyi pot, veselce, etc.). We see the prospects for further research in the continuation of work on the compilation of the Dictionary of Podillia Dialects.

Whole Genome Analysis in Consanguineous Families Reveals New Loci for Speech Sound Disorder (SSD).

Speech is the most common means of communication in humans. Any defect in accurate speech production ability results in the development of speech sound disorder (SSD), a condition that can significantly impair an individual's academic performance, social interactions, and relationships with peers and adults. This study investigated the genetic basis of SSD in three Pakistani families. We performed family-based genome-wide parametric linkage analysis and homozygosity mapping in three consanguineous families with SSD from the Punjab province of Pakistan. The Test for Assessment of Articulation and Phonology in Urdu (TAAPU) was used to analyze the speech articulation data and determine the Percentage Correct Consonants (PCC) score. The PCC score defined the affected and unaffected individuals in each family. Parametric linkage analysis revealed a linkage to chromosome 5 (5q21.3-5q23.1) with a significant logarithm of the odds (LOD) score of 3.13 in a Pakistani family with specific language impairment-97 (PKSLI-97) under an autosomal recessive mode of inheritance. The other two families showed a suggestive linkage at 6p22.1, 14q12, and 16q12.1 under the recessive mode of inheritance. Interestingly, homozygosity mapping showed a loss of heterozygosity in the linkage region at 5q15-5q23.1, shared among seven affected (mostly in the younger generation) and one unaffected individual of PKSLI-97. Our analysis identified the 6p22 locus previously implicated in dyslexia, childhood apraxia of speech (CAS), and language impairment, confirming the role of KIAA0319 and DCDC2 in this locus. These findings provide statistical evidence for the genomic regions associated with articulation disorder and offer future opportunities to further the role of genes in speech production.

Refining maritime Automatic Speech Recognition by leveraging synthetic speech

Maritime transport serves as a critical component of global trade and logistics, enabling the movement of goods and resources across oceans and waterways. Especially in busy waterways and ports, effective and accurate communication is essential, as it ensures the seamless exchange of information and the coordinated execution of port activities. However, comprehensibility is often hindered by factors such as poor audio quality, background noise, and diverse languages and accents. Automatic Speech Recognition (ASR) systems can mitigate these issues by providing real-time transcription and enabling the implementation of automated, value-adding services to enhance situational awareness. While pre-trained ASR models excel on general speech, maritime ASR faces unique challenges due to a lack of annotated data, diverse accents, and specialized terminology.To this end, we focus on improving the transcription quality of pre-trained ASR models for maritime communication with a particular focus on accurately recognizing maritime-specific terminology such as vessel and location names. Due to the scarcity of transcribed maritime communication, we create a synthetic training dataset tailored to regional maritime terminology. The synthetic audio is augmented with general human speech and used to fine-tune an end-to-end ASR model under various settings. The evaluation of the models employs a proprietary dataset of regional maritime radio communication from the port of Hamburg.The experimental results demonstrate a notable enhancement in ASR performance. Specifically, our approach yields an absolute improvement over the pre-trained baseline of 13.46% Word-Error-Rate and an increase of 41.57% recall for vessel names and 38.65% recall for locations. Our findings underscore the efficacy of integrating synthetic training data to address the challenges encountered in maritime ASR, paving the way for more robust and accurate speech recognition systems tailored to maritime applications.

Association between tooth loss-related speech and psychosocial impairment with cognitive function: A pilot study in Hong Kong's older population.

Tooth loss has been associated with cognitive decline, but the underlying mechanisms involving speech and psychosocial impairment remain unclear. To investigate the impact of tooth loss-related speech and psychosocial impairment on cognitive function in Hong Kong's older population. Seventy-six Cantonese-speaking participants between the ages of 51-92 were classified into three groups: patients with complete dentures (CD), partially edentulous patients with less than 10 occluding tooth pairs (OU <10), and at least 10 occluding tooth pairs (OU ≥10). Cognitive function was assessed using the Montreal Cognitive Assessment Hong Kong Version, One-minute Verbal Fluency Task and Hayling Sentence Completion Test. Objective and subjective speech assessments were carried out using artificial intelligence speech recognition algorithm and a self-designed speech questionnaire. The impact of tooth loss on psychosocial condition was evaluated by the Reading the Mind in the Eyes Test and a self-designed questionnaire. Statistical analyses (one-way ANOVA, ANCOVA, Kruskal-Wallis test, Spearman correlation test) were performed. Tooth loss was significantly associated with lower cognitive function (p = .008), speech accuracy (p = .018) and verbal fluency (p = .001). Correlations were found between cognitive function and speech accuracy (p < .0001). No significant difference in tooth loss-related psychosocial impact was found between the three groups. While warranting larger sample sizes, this pilot study highlights the need for further research on the role of speech in the association between tooth loss and cognitive function. The potential cognitive impact of tooth retention, together with its known biological and proprioceptive benefits, supports the preservation of the natural dentition.

Integrating international Chinese visualization teaching and vocational skills training: leveraging attention-connectionist temporal classification models.

The teaching of Chinese as a second language has become increasingly crucial for promoting cross-cultural exchange and mutual learning worldwide. However, traditional approaches to international Chinese language teaching have limitations that hinder their effectiveness, such as outdated teaching materials, lack of qualified instructors, and limited access to learning facilities. To overcome these challenges, it is imperative to develop intelligent and visually engaging methods for teaching international Chinese language learners. In this article, we propose leveraging speech recognition technology within artificial intelligence to create an oral assistance platform that provides visualized pinyin-formatted feedback to learners. Additionally, this system can identify accent errors and provide vocational skills training to improve learners' communication abilities. To achieve this, we propose the Attention-Connectionist Temporal Classification (CTC) model, which utilizes a specific temporal convolutional neural network to capture the location information necessary for accurate speech recognition. Our experimental results demonstrate that this model outperforms similar approaches, with significant reductions in error rates for both validation and test sets, compared with the original Attention model, Claim, Evidence, Reasoning (CER) is reduced by 0.67%. Overall, our proposed approach has significant potential for enhancing the efficiency and effectiveness of vocational skills training for international Chinese language learners.

A Novel Dual Kernel Support Vector-Based Levy Dung Beetle Algorithm for Accurate Speech Emotion Detection

A Novel Dual Kernel Support Vector-Based Levy Dung Beetle Algorithm for Accurate Speech Emotion Detection

Implementation of Data Mining for Speech Recognition Classification of Sundanese Dialect Using KNN Method with MFCC Feature Extraction

The importance of preservation and development of speech recognition technology for regional languages such as Sundanese, which have unique phonetic characteristics. Regional language speech recognition can assist in the development of local, educational, and cultural preservation applications to implement and evaluate the effectiveness of the combination of MFCC and KNN methods in classifying Sundanese dialect speech recognition. Methods used include trait extraction with MFCC, which converts voice data into numerical representations based on frequency characteristics, and classification with KNN, which groups data based on similarity to train data. The Dataset used consisted of speech recordings of Western and Southern Sundanese dialects. The results showed that the k-Nearest Neighbors (KNN) method can classify Sundanese dialect speech recognition with an accuracy of 80.00%, showing good ability in distinguishing "Western" and "southern" dialects. Mel-Frequency Cepstral Coefficients (MFCC) proved to be very effective in extracting sound features, helping KNN achieve low error rates. The combination of MFCC and KNN proved effective for speech recognition classification of Sundanese dialects, providing satisfactory results with high accuracy.

Low Resource Malay Dialect Automatic Speech Recognition Modeling Using Transfer Learning from a Standard Malay Model

Approaches to automatic speech recognition have transited from Hidden Markov Model (HMM)-based ASR to deep neural networks. The advantages of deep neural network approaches are that they can be developed quickly and perform better given large language resources. Nevertheless, dialect speech recognition is still challenging due to the limited resources. Transfer learning approaches have been proposed to improve speech recognition for low resources. In the first approach, the model is pre-trained on a large and diverse labeled dataset to learn the acoustic and language patterns from the speech signal. Then, the model parameters are updated with a new dataset, and the pre-trained model is fine-tuned on a low-resource language dataset. The fine-tuning process is usually completed by freezing the pre-trained layers and training the remaining layers of the model on the low-resource language corpus. Another approach is to use a pre-trained model to capture the compact and meaningful features as input to the encoder. Pre-training in this approach usually involves using unsupervised learning methods to train models on a corpus of large amounts of unmarked data. It enables the model to learn the general patterns and relationships between the input speech signals. This paper proposes a training recipe using transfer learning and Standard Malay models to improve automatic speech recognition for Kelantan and Sarawak Malay dialects.