Exploring Universal Singing Speech Language Identification Using Self-Supervised Learning Based Front-End Features

Language identification technology is widely used in the domains of machine learning and text mining. Many researchers have achieved excellent results on a few selected European languages. However, the majority of African and Asian languages remain untested. The primary objective of this research is to evaluate the performance of our new n‑gram based language identification algorithm on 68 written languages used in the European, African and Asian regions. The secondary objective is to evaluate how n‑gram orders and a mix n‑gram model affect the relative performance and accuracy of language identification. The n-gram based algorithm used in this paper does not depend on the n‑gram frequency. Instead, the algorithm is based on a Boolean method to determine the output of matching target n‑grams to training n‑grams. The algorithm is designed to automatically detect the language, script and character encoding scheme of a written text. It is important to identify these three properties due to the reason that a language can be written in different types of scripts and encoded with different types of character encoding schemes. The experimental results show that in one test the algorithm achieved up to 99.59% correct identification rate on selected languages. The results also show that the performance of language identification can be improved by using a mix n‑gram model of bigram and trigram. The mix n-gram model consumed less disk space and computing time, compared to a trigram model. DOI: 10.4038/icter.v2i2.1385 The International Journal on Advances in ICT for Emerging Regions 2009 02 (02): 21-28

Speech, languages, and background voices are overlapped in the multi-speaker environment. The speech separation process is necessary for downstream applications like language and speaker identification and ASR tasks. A combined speech separation and language recognition approach is proposed in this paper. First, the monolingual information is extracted from a multi-speaker multilingual speech scenario, then fed the separated speech to a language identification model. The speech separation model is trained in a single-channel overlapped speech data, and the language identification model is trained in Oriental language corpus. The combined effect of these two models is evaluated in the task of language identification. When the language recognition network is fed with the audio spectrograms separated by the source separation network, the Chinese and Korean language recognition results are significantly improved in terms of accuracy and recall rate than the mixed audio spectrograms.

The availability of huge annotated datasets and deep learning have led to astounding Natural Language Processing (NLP) progress. However, there is a plethora of low-resource languages, lack the labelled corpora to train the models in the supervision. This kind of digital divide does not allow fair NLP development across language groups. Self-supervised learning (SSL) is a new paradigm in recent years that leverages large amounts of untagged text to learn powerful representations with little or no manual annotation. This paper is a review of self-supervised deep learning models in low-resource NLP tasks. It begins with the definition of the principles of the SSL and the difference between this approach and supervised and unsupervised approaches. We describe why methods such as masked language modeling, contrastive learning, and autoregressive modeling underlie modern pre-trained transformers such as BERT, GPT, and mBERT. A particular interest is paid to multilingual and cross lingual SSL schemes which allow knowledge transfer between high-resource and low resource languages. Low-resource tasks, including machine translation, sentiment analysis, speech-to-text, and information retrieval are reviewed. As per benchmark studies, even in small labelled samples, the SSL models can achieve big gains in accuracy and generalization. Computation cost, representational bias and morphologically rich and under-documented language evaluation, are however, problematic. The researchers observe that the study suffers certain limitations, such as over-reliance on high-resource pretraining information, inequity between linguistic groups, and the difficulty of deploying large-scale SSL models in resource-constrained circumstances. The future directions include lightweight multilingual models, federation of learning in NLP and symbolic linguistic knowledge and combination with SSL. Self-supervised deep learning bridges an essential gap between the high-resource and low-resource languages, and would be a highly valuable step toward inclusive, global NLP innovation.

Early identification of speech, language, and literacy difficulties in preschool-aged children is crucial for timely intervention and prevention of long-term academic challenges. This study aimed to evaluate the Speech Language and Early Literacy Screening Checklist (SLELSC), a tool designed to assess key developmental milestones in 5- to 6-year-old children before they enter elementary school. The checklist was developed based on critical aspects of language acquisition, narrative discourse, phonological development, and emergent literacy skills, which collectively support the transition to formal education. A total of 155 children participated in this study, including 45 typically developing children and 110 children identified as at risk for speech, language, or literacy difficulties. The SLELSC demonstrated strong content validity (average CVI = 0.91) and clinical validity, with a significant difference in scores between the at-risk and typically developing groups. ROC curve analysis identified an optimal cut-off score of 19.5, yielding a sensitivity of 84.4% and specificity of 76.4%. The checklist also showed high diagnostic accuracy, with a positive predictive value of 92.31% and a comparatively lower negative predictive value of 59.38%, indicating that while the tool is effective in identifying children at risk, further refinement may be needed to improve its ability to rule out difficulties. The findings of this study provide strong evidence for the SLELSC as a valid and reliable tool for use in clinical and educational settings. By facilitating early assessment, this checklist contributes to the timely provision of support services, helping to mitigate the long-term effects of language and literacy delays. What is already known on this subject Early identification of speech, language, and literacy difficulties is essential for academic and social development. Approximately 10%-20% of preschoolers experience such challenges, which can lead to long-term educational and psychosocial difficulties. While various developmental screening tools exist, they often lack a specific focus on key linguistic domains for children nearing school entry. In South Korea, there is no standardised screening tool for 5- to 6-year-olds. The Speech Language and Early Literacy Screening Checklist was developed to address this gap, demonstrating initial reliability and validity, but further validation is needed to confirm its diagnostic accuracy.What this study adds to existing knowledge This study confirms the Speech Language and Early Literacy Screening Checklist as a valid and reliable tool for identifying preschool-aged children at risk for language and literacy difficulties. The checklist demonstrated strong clinical accuracy, with high sensitivity and specificity, making it effective for early detection. Additionally, it highlights the need for domain-specific cut-off scores to enhance diagnostic precision. By facilitating timely intervention before school entry, this screening tool provides a practical solution for educators and clinicians, helping to mitigate long-term academic and social challenges associated with early speech and language delays.What are the potential or actual clinical implication of this work? The Speech Language and Early Literacy Screening Checklist provides a clinically valid and reliable tool for early identification of at-risk children before school entry. Its high sensitivity ensures that most children with speech, language, or literacy difficulties receive timely intervention. The checklist can be effectively utilised in preschool settings by educators and clinicians to support early screening and referral. Additionally, refining cut-off scores for specific domains may further improve diagnostic accuracy. Future research should explore its application across diverse populations and its long-term impact on academic and social outcomes, strengthening early intervention strategies in clinical and educational settings.

In recent years, with the rise of human-computer interaction and the successful application of end-to-end models in the field of speech recognition, the construction of end-to-end speech recognition models has received extensive attention. Relying on the multi-task learning method and the connection between language identification and speech recognition, we proposed an end-to-end Transformer model, which is a multilingual speech recognition model integrating language identification. The model takes the speech recognition task as the main task and the language identification task as the auxiliary task. In this paper, the validity of the model is verified by using the datasets of 13 languages in the 2021 Oriental Language Recognition challenge (OLR). The experimental results show that the model constructed in this paper has a relative improvement of 37.46% in the speech recognition task compared with the baseline system proposed by the OLR organizer. The accuracy of language identification reaches 89.70 %. The results can get the fifth place in the 2021 OLR constraint track of speech recognition equally.

Language identification is the task of automatically determining the identity of a language conveyed by a spoken segment. It has a profound impact on the multilingual interoperability of an intelligent speech system. Despite language identification attaining high accuracy on medium or long utterances (>3 s), the performance on short utterances (<=1s) is still far from satisfactory. We propose a BERT-based language identification system (BERT-LID) to improve language identification performance, especially on short-duration speech segments. We extend the original BERT model by taking the phonetic posteriorgrams (PPG) derived from the front-end phone recognizer as input. Then we deployed the optimal deep classifier followed by it for language identification. Our BERT-LID model can improve the baseline accuracy by about 6.5% on long-segment identification and 19.9% on short-segment identification, demonstrating our BERT-LID’s effectiveness to language identification.

This study proposes a novel deep neural network (DNN) based approach to language identification (LID) for the NIST 2015 Language Recognition (LRE) i-Vector Machine Learning Challenge. State-of-the-art DNN based LID systems utilize large amounts of labeled training data. The 2015 LRE i-Vector Machine Learning Challenge limits the access to only ready-to-use i-Vectors for LID system training and testing. This poses unique challenges in designing DNN based LID systems, since optimized front-ends and network architectures can no longer be used. We propose to use the training i-Vectors to train an initial DNN for LID. Next, we present a novel strategy to use this initial DNN to estimate out-of-set language labels from the development data. The final DNN for LID is trained using the original training data, and the estimated out-of-set language data. We show that augmenting the training set with out-of-set labels leads to significant improvement in the LID performance. Our approach obtains very competitive costs (defined by NIST) of 26.56, and 25.98 respectively, on the progress and evaluation subsets of the challenge. Since the amount of training data is very limited (300 i-Vectors per language), this study outlines a successful recipe for DNN based LID using very limited resources.

Language Identification of Intra-Word Code-Switching for Arabic–English

PurposeWith the rapid emergence and explosion of the internet and the trend of globalization, a tremendous number of textual documents written in different languages are electronically accessible online from the world wide web. Efficiently and effectively managing these documents written in different languages is important to organizations and individuals. Therefore, the purpose of this paper is to propose letter frequency neural networks to enhance the performance of language identification.Design/methodology/approachInitially, the paper analyzes the feasibility of using a windowing algorithm in order to find the best method in selecting the features of Arabic script documents language identification using backpropagation neural networks. Previously, it had been found that the sliding window and non‐sliding window algorithm used as feature selection methods in the experiments did not yield a good result. Therefore, this paper proposes, a language identification of Arabic script documents based on letter frequency using a backpropagation neural network and used the datasets belonging to Arabic, Persian, Urdu and Pashto language documents which are all Arabic script languages.FindingsThe experiments have shown that the average root mean squared error of Arabic script document language identification based on letter frequency feature selection algorithm is lower than the windowing algorithm.Originality/valueThis paper highlights the fact that using neural networks with proper feature selection methods will increase the performance of language identification.

It is critical to have efficient automated detection systems in place since the proliferation of hate speech and inflammatory language on social media platforms has been accelerated by their fast development. Attempts to capture contextual connections were a challenge for traditional machine learning methods like SVM and Logistic Regression, which depended on manually created features. Though they mostly handled text as sequences and neglected underlying relationship structures, deep learning and Transformer-based models learned contextual embeddings, which enhanced performance. To overcome this shortcoming, this research presents a GNN-based framework for modeling textual data as graphs, which allows for a more comprehensive portrayal of the semantic and relational relationships between words and texts. For experimental assessment, three datasets were used: Davidson, which had 24,783 tweets with 5.8% hate, 77.4% offensive, and 16.8% neither; HASOC, which contained around 9,000-12,000 tweets in English and Hindi, with a somewhat balanced distribution; and Founta, which contained 80,000+ tweets in all three categories (hate, abusive, and normal). Baselines such as Logistic Regression, SVM, CNN, BiLSTM, and BERT classifiers were tested against the suggested architecture, which combines GCN, GAT, and GraphSAGE with pre-trained embeddings (GloVe and BERT). Dataset after dataset shows that GNN-based models perform far better than baselines. The F1-scores of GraphSAGE and BERT on the Davidson dataset were 91.7% and 87.8%, respectively. For the HASOC dataset, GraphSAGE achieved an F1-score of 89.9%, which was higher than BERT's 85.5%. Again, GraphSAGE topped the Founta dataset with a 90.2% F1-score, well surpassing BERT's 86.1% performance. Compared to the most advanced Transformer models, these findings show an improvement of 3-5% in F1-score. The results show that abusive language and hate speech identification are better handled by graph-based representations, which give a more complete picture of the linguistic and relational environment. This study lays the groundwork for future research on multimodal, multilingual, and real-time detection systems, and it also sets GNNs as a strong and scalable method for online content moderation. Keywords: Handwritten Hate Speech, Offensive Language, Graph Neural Networks (GNNs), GCN, GAT, GraphSAGE, Social Media, Text Classification,

Training deep learning models on medical images heavily depends on experts’ expensive and laborious manual labels. In addition, these images, labels, and even models themselves are not widely publicly accessible and suffer from various kinds of bias and imbalances. In this paper, chest X-ray pre-trained model via self-supervised contrastive learning (CheSS) was proposed to learn models with various representations in chest radiographs (CXRs). Our contribution is a publicly accessible pretrained model trained with a 4.8-M CXR dataset using self-supervised learning with a contrastive learning and its validation with various kinds of downstream tasks including classification on the 6-class diseases in internal dataset, diseases classification in CheXpert, bone suppression, and nodule generation. When compared to a scratch model, on the 6-class classification test dataset, we achieved 28.5% increase in accuracy. On the CheXpert dataset, we achieved 1.3% increase in mean area under the receiver operating characteristic curve on the full dataset and 11.4% increase only using 1% data in stress test manner. On bone suppression with perceptual loss, we achieved improvement in peak signal to noise ratio from 34.99 to 37.77, structural similarity index measure from 0.976 to 0.977, and root-square-mean error from 4.410 to 3.301 when compared to ImageNet pretrained model. Finally, on nodule generation, we achieved improvement in Fréchet inception distance from 24.06 to 17.07. Our study showed the decent transferability of CheSS weights. CheSS weights can help researchers overcome data imbalance, data shortage, and inaccessibility of medical image datasets. CheSS weight is available at https://github.com/mi2rl/CheSS.

Phonotactic modelling, typically in the form of a PPRLM system, forms a key component in state-of-the-art Language Identification (LID) systems. Given the objective of PPRLM systems is to capture as accurately as possible the phonotactics which characterise a language, it is assumed that the minimisation of Phone Error Rate (PER) is a precursor to achieving this effectively. In this paper we examine the relevance of PER as a metric for determining eventual LID performance. In order to conduct this investigation we make use of the CallHome corpus, based on the premise it provides a better representation for the style of discourse and channel conditions encountered in the Conversational Telephone Speech (CTS), which is now the focus of current NIST LID evaluations. Using CallHome instead of the OGI-MLTS corpus to train phone recognisers, we obtained significantly improved results, with an average improvement of approximately 6% absolute across the 30, 10 and 3 seconds tasks for the NIST 1996 and 2003 evaluations. We also examine the impact of tuning the individual front-end recognisers, on both the resultant PER of other languages and against the resultant LID performance. We find that PER has a number of limitations in indicating both the degree and direction of changes to LID performance. Accordingly, we propose a new metric which is better suited for forecasting the impact on LID performance when the phone recogniser front-end is modified.

This article addresses language identification at the word level in Indian social media corpora taken from Facebook, Twitter and WhatsApp posts that exhibit code-mixing between English-Hindi, English-Bengali, as well as a blend of both language pairs. Code-mixing is a fusion of multiple languages previously mainly associated with spoken language, but which social media users also deploy when communicating in ways that tend to be rather casual. The coarse nature of code-mixed social media text makes language identification challenging. Here, the performance of deep learning on this task is compared to feature-based learning, with two Recursive Neural Network techniques, Long Short Term Memory (LSTM) and bidirectional LSTM, being contrasted to a Conditional Random Fields (CRF) classifier. The results show the deep learners outscoring the CRF, with the bidirectional LSTM demonstrating the best language identification performance.

Using more than language in a single discourse is a practical phenomenon appears in bilingual and multilingual communities known as mixed speech communication. The mixed speech occurs in communication to express ideas and thoughts using vocabulary of all used languages. This paper addresses the problem of automatic speech recognition and language identification for Sudanese Arabic and English languages in mixed speech mode, in which number and location of languages are not previously known. Native language in mixed speech is dominant based on assumption that native speaker does not suddenly reconfigure his articulation organs to produce sounds when switching to other language. This study proposes a generalized framework of Automatic Speech Recognin in Mixed Speech mode (ASR-MS). Proposed framework defines mixed speech as a hybrid language not belong to any language participates in mixed sentence. This new language is processed based on its distinct attributes not based on attributes of languages originated from. A single Language Model (AM) is built based on Sudanese Arabic pronunciation. The supporting components such as phonetic dictionary and language lexicon are bilingual, keeps words in their original forms. For measuring the soundness of a framework, a 100 Sudanese Arabic – English daily life mixed sentences were collected and recorded. Preliminary results are encouraging towards building a generalized mixed speech recognition and language identification based on speakers instead of languages.

In this paper, phoneme sequences are used as language information to perform code-switched language identification (LID). With the one-pass recognition system, the spoken sounds are converted into phonetically arranged sequences of sounds. The acoustic models are robust enough to handle multiple languages when emulating multiple hidden Markov models (HMMs). To determine the phoneme similarity among our target languages, we reported two methods of phoneme mapping. Statistical phoneme-based bigram language models (LM) are integrated into speech decoding to eliminate possible phone mismatches. The supervised support vector machine (SVM) is used to learn to recognize the phonetic information of mixed-language speech based on recognized phone sequences. As the back-end decision is taken by an SVM, the likelihood scores of segments with monolingual phone occurrence are used to classify language identity. The speech corpus was tested on Sepedi and English languages that are often mixed. Our system is evaluated by measuring both the ASR performance and the LID performance separately. The systems have obtained a promising ASR accuracy with data-driven phone merging approach modelled using 16 Gaussian mixtures per state. In code-switched speech and monolingual speech segments respectively, the proposed systems achieved an acceptable ASR and LID accuracy.