Dialect Identification Research Articles

Dialetos e Variações Linguísticas do Hebraico Norte-Israelita

O livro de Juízes condensa tradições regionais por meio de um rico intercâmbio de variações linguísticas que pode ser resgatado pela dialetologia. Este artigo teve por objetivo identificar, catalogar e analisar alguns dos múltiplos dialetos existentes no livro de Juízes. O avanço das pesquisas na área da linguística hebraica tem aberto caminhos para a identificação de dialetos, localização e datação de textos da Bíblia Hebraica, como é o caso das obras de Gary A. Rendsburg, William Schniedewind, W. Randall Garr, Ran Zadok, Nadav Naaman e Tania Notarius, os quais formam a base teórica desta pesquisa. O procedimento metodológico adotado aqui seguiu o método da pesquisa de Gary Rendsburg: distribuição, evidências extra bíblicas, oposição e concentração. A hipótese desta pesquisa propõe a existência de dialetos hebraicos norte-israelitas registrados nos textos do livro de Juízes, os quais podem ser identificados, catalogados e analisados seguindo os métodos desta pesquisa.

Multitask Learning with Local Attention for Tibetan Speech Recognition

In this paper, we propose to incorporate the local attention in WaveNet-CTC to improve the performance of Tibetan speech recognition in multitask learning. With an increase in task number, such as simultaneous Tibetan speech content recognition, dialect identification, and speaker recognition, the accuracy rate of a single WaveNet-CTC decreases on speech recognition. Inspired by the attention mechanism, we introduce the local attention to automatically tune the weights of feature frames in a window and pay different attention on context information for multitask learning. The experimental results show that our method improves the accuracies of speech recognition for all Tibetan dialects in three-task learning, compared with the baseline model. Furthermore, our method significantly improves the accuracy for low-resource dialect by 5.11% against the specific-dialect model.

Attention-based CNN-BiLSTM for Dialect Identification on Javanese Text

This study proposes a hybrid deep learning models called attention-based CNN-BiLSTM (ACBiL) for dialect identification on Javanese text. Our ACBiL model comprises of input layer, convolution layer, max pooling layer, batch normalization layer, bidirectional LSTM layer, attention layer, fully connected layer and softmax layer. In the attention layer, we applied a hierarchical attention networks using word and sentence level attention to observe the level of importance from the content. As comparison, we also experimented with other several classical machine learning and deep learning approaches. Among the classical machine learning, the Linear Regression with unigram achieved the best performance with average accuracy of 0.9647. In addition, our observation with the deep learning models outperformed the traditional machine learning models significantly. Our experiments showed that the ACBiL architecture achieved the best performance among the other deep learning methods with the accuracy of 0.9944.

Natural language processing for similar languages, varieties, and dialects: A survey

AbstractThere has been a lot of recent interest in the natural language processing (NLP) community in the computational processing of language varieties and dialects, with the aim to improve the performance of applications such as machine translation, speech recognition, and dialogue systems. Here, we attempt to survey this growing field of research, with focus on computational methods for processing similar languages, varieties, and dialects. In particular, we discuss the most important challenges when dealing with diatopic language variation, and we present some of the available datasets, the process of data collection, and the most common data collection strategies used to compile datasets for similar languages, varieties, and dialects. We further present a number of studies on computational methods developed and/or adapted for preprocessing, normalization, part-of-speech tagging, and parsing similar languages, language varieties, and dialects. Finally, we discuss relevant applications such as language and dialect identification and machine translation for closely related languages, language varieties, and dialects.

A Deep Learning Approach for the Romanized Tunisian Dialect Identification

Language identification is an important task in natural language processing that consists in determining the language of a given text. It has increasingly picked the interest of researchers for the past few years, especially for code-switching informal textual content. In this paper, we focus on the identification of the Romanized user-generated Tunisian dialect on the social web. We segment and annotate a corpus extracted from social media and propose a deep learning approach for the identification task. We use a Bidirectional Long Short-Term Memory neural network with Conditional Random Fields decoding (BLSTM-CRF). For word embeddings, we combine word-character BLSTM vector representation and Fast Text embeddings that takes into consideration character n-gram features. The overall accuracy obtained is 98.65%.

Information about talker dialect is available in high-frequency region

Given that almost all cues to speech intelligibility are contained within the low-frequency region, high-pass filtering has not been used in speech perception research as often as low-pass filtering. We provide evidence from high-pass filtered speech that information about talker dialect is available in the high-frequency region even in the absence of intelligibility cues. Setting the upper frequency limit at 11 kHz, sentences edited out of spontaneous conversations of 20 talkers from Ohio and 20 from North Carolina were high-pass filtered with frequency cut-offs varying from 0.7 to 5.56 kHz and presented to listeners from Ohio. Results showed that listeners were still sensitive to differences between the two dialects at the two highest cut-offs, 3.32 kHz and 5.56 kHz, and dialect identification was mediated by talker sex. Also, identification of talker sex was affected by dialect, with Ohio variety providing more cues to talker sex at 3.32 kHz and the variety in North Carolina at 5.56 kHz. Speech intelligibility was reduced above 3 kHz, and Ohio speakers were more intelligible overall. Irrespective of the intelligibility loss, the results suggest that residual dialect cues are still distributed in the high frequency region, and are preserved differently in male and female voices.

Categorization of U.S. regional dialects and race from speech

Listeners extract a range of indexical information about speakers from the acoustic speech signal. Listeners show moderate to high accuracy with the two dimensions investigated here—race and regional dialect—with generally higher accuracy for race than regional dialect. In this project, both race and regional dialect identification from the same talkers were tested when the dimensions were varied orthogonally. In a forced-choice categorization task, listeners were presented with 144 unique audio clips of monolingual English talkers from Midland, New York City, and Southern dialect regions who were either Asian, Black, or white. Stimuli were presented to listeners in two separate, counter-balanced testing blocks (i.e., one for race and one for regional dialect). Results showed higher accuracy for race identification than regional dialect, with white Midland talkers most accurately identified by race and Southern Black talkers most accurately identified by dialect. Both tasks showed significant interactions between the two indexical dimensions. The results suggest that an irrelevant indexical dimension can impact perception of the relevant indexical dimension. Further, the composite of a listener's identity and their experiences with people of various ethnicities, racial and dialect groups can shape their perception and response biases. [Work supported by Indiana University Hutton Honors College.]

Frameworks of Reference in the Identification of Latin Dialects

Various studies proved that the methodology of J. Herman produces plausible and verifiable results in the field of Latin dialectology, but certain methodological questions remained still unanswered regarding our points of reference in the decision which proportions of the data of the inscriptional faults are classified significant; how to decide on the basis of the proportion of a certain error type if a certain linguistic change was in progress, if it was completed, or if it was not active in the examined territory; which types of errors can serve as base or bases of comparison for a specific examined error type; which periods and territories should be the point of reference in comparisons. In the present study, we attempt to give answers to these questions by running statistical surveys using different points of reference in statistical significance and different bases of comparisons in the error types, and we set up a list of expected results based on the known tendencies of sound changes in Vulgar Latin against which we will measure the actual results of the survey in order to determine which methods were the most effective in meeting the expected picture that we already know about the development of Romance languages.

Vowels and tones as acoustic cues in Chinese subregional dialect identification

The aim of the present perceptual study is to weight tones and vowels as acoustic cues in Chinese subregional dialect identification, and to test the credibility of the subregional dialect classification that has been proposed in the literature. Our findings show that listeners are able to pinpoint speakers’ subregional dialect even when only given monosyllabic Chinese word stimuli, either natural or tone-transplanted. The results agree with the impressionistic claim that both vowels and tones contribute to perceptual subregional dialect identification. However, vowel quality differences make a greater contribution than the tone differences – which contradicts the order of importance predicted in the impressionistic literature. Strong interactions between vowels and tones are also found.

Automatic accent identification as an analytical tool for accent robust automatic speech recognition

We present a novel study of relationships between automatic accent identification (AID) and accent-robust automatic speech recognition (ASR), using i-vector based AID and deep neural network, hidden Markov Model (DNN-HMM) based ASR. A visualization of the AID i-vector space and a novel analysis of the accent content of the WSJCAM0 corpus are presented. Accents that occur at the periphery of AID space are referred to as “extreme”. We demonstrate a negative correlation, with respect to accent, between AID and ASR accuracy, where extreme accents exhibit the highest AID and lowest ASR performance. These relationships between accents inform a set of ASR experiments in which a generic training set (WSJCAM0) is supplemented with a fixed amount of accented data from the ABI (Accents of the British Isles) corpus. The best performance across all accents, a 32% relative reduction in errors compared with the baseline ASR system, is obtained when the supplementary data comprises extreme accented speech, even though this accent accounts for just 14% of the test data. We conclude that i-vector based AID analysis provides a principled approach to the selection of training material for accent robust ASR. We speculate that this may generalize to other detection technologies and other types of variability, such as Speaker Identification (SI) and speaker variability.

Machine learning applied in natural language processing

Machine learning applied in natural language processing

Spoken Arabic dialect recognition using X-vectors

AbstractThis paper describes our automatic dialect identification system for recognizing four major Arabic dialects, as well as Modern Standard Arabic. We adapted the X-vector framework, which was originally developed for speaker recognition, to the task of Arabic dialect identification (ADI). The training and development ADI VarDial 2018 and VarDial 2017 were used to train and test all of our ADI systems. In addition to the introduced X-vectors, other systems use the traditional i-vectors, bottleneck features, phonetic features, words transcriptions, and GMM-tokens. X-vectors achieved good performance (0.687) on the ADI 2018 Discriminating between Similar Languages shared task testing dataset, outperforming other systems. The performance of the X-vector system is slightly improved (0.697) when fused with i-vectors, bottleneck features, and word uni-gram features.

Identification of regional dialects of Telugu language using text independent speech processing models

Telugu language is one of the important languages in the world. The language that is spoken by most of the people in a region is called as dialect. In the recent days, speech recognition system is present in almost all electronic devices. In this, dialects of particular language perform a vital role. The accurate dialects identification technique helps in not only enhancing its features but also expected to provide in modern services in health and telemedicine for older and homebound peoples. Like any other language, even Telugu language has diversified itself into different dialects viz., Telangana, Kostha Andhra, and Rayalaseema. Combination of all the dialects is the language TELUGU and it is a perfect blend of elegance in Sanskrit, sweetness in Tamil along with the essence of Kannada language. The formation of dialects can be of different reasons. For speech processing research, till today there is no standard speech database created for Telugu dialects. In this paper we developed a speech database that can be utilized for the recognition of Telugu dialects and we had applied two modeling techniques that are, Hidden Markov Model (HMM) and Gaussian mixture model (GMM) in order to recognize the dialects of Telugu language by using speech independant utterances. We imposed Mel-Frequency Cepstral Coefficient for extracting the spectral features from the obtained speech data and observed that GMM provides better accurate results than HMM.

Mel-Weighted Single Frequency Filtering Spectrogram for Dialect Identification

In this study, we propose Mel-weighted single frequency filtering (SFF) spectrograms for dialect identification. The spectrum derived using SFF has high spectral resolution for harmonics and resonances while simultaneously maintaining good time-resolution of some speech excitation features such as impulse-like events. The SFF spectrum can represent speech characteristics such as burst time and glottal closure instances better than the short-time Fourier transform (STFT) spectrum. Our hypothesis is that these intricate representations in the SFF spectrum should help in distinguishing dialects. Therefore, we built a dialect identification system which uses an unsupervised, bottleneck feature representation of the Mel-weighted SFF spectrogram (Mel-SFF spectrogram) with sequence-to-sequence deep autoencoders. The language invariance of the proposed system was evaluated using two datasets: the UT-Podcast database (English) and the STYRIALECT database (German). The proposed representations gave a relative improvement of 9.47% and 4.69% in unweighted average recall (UAR) compared to the best baseline method on the development and test datasets, respectively, of the UT-Podcast database. The proposed representations also gave a comparable performance to the best baseline method for the STYRIALECT database. In addition, the fusion of the autoencoder bottleneck features computed from the Mel-SFF and Mel-STFT spectrograms improved the overall performance indicating complementary information between these features. By further analyzing the performance of the proposed representation with different utterance lengths using the UT-Podcast database, we observed that the proposed representation performed better on short utterances. The improved performance given by the Mel-weighted SFF spectrogram for recognizing dialects in both databases supports our hypothesis.

Accent Identification of Ethnically Diverse Nigerian English Speakers

It is imperative to improve the speech recognition system as human-machine interfaces are advancing in the growing global market of technologies. There are quite a number of Nigerian English speakers’ accents to which the speech recognition systems are not sufficiently exposed. Accents may suggest a lot of information about someone’s whereabouts, for example, their native language, place of origin, or ethnic groups and accent classification. Given the importance of accents, efficiency and accuracy of speech recognition systems can be improved with training data of diverse accents. This research provides support for accent-dependent automatic speech recognition by deploying a supervised learning algorithm to the task of recognizing three Nigerian ethnic groups (Yoruba, Igbo, and Hausa) and distinguish them based on their accents by constructing sequential Mel-Frequency Cepstral Coefficients (MFCC) features from the frames of the audio sample. Our results show that concatenating the MFCC features sequentially and applying a supervised learning technique to provide a solution to the problem of identifying and classifying accents works efficiently and accurately.

An open speech resource for Tibetan multi-dialect and multitask recognition

This paper introduces a Tibetan multi-dialect data resource for multitask speech research. It can be used for Tibetan multi-dialect speech recognition, Tibetan speaker recognition, Tibetan dialect identification, and Tibetan speech synthesis. The resource consists of 30 hours Lhasa-U-Tsang dialect; 8.7 hours Kham dialect, including 3.4 hours Yushu dialect, 3.3 hours Dege dialect and 2 hours Changdu dialect; 10 hours Amdo pastoral dialect. Other resources are also provided for Lhasa-U-Tsang dialect including phoneme set, pronunciation dictionary and the codes for constructing the Lhasa-U-Tsang speech recognition baseline system. Meanwhile, for Tibetan multi-dialect and multitask speech recognition, the codes and recognition results based on WaveNet-connectionist temporal classification (WaveNet-CTC) are provided. All the resources are free for researchers and publicly available, which effectively compensates for the shortage of public Tibetan multi-dialect speech resources in order to promote the development of Tibetan multi-dialect speech processing technology.

Automatic dialect identification system for Kannada language using single and ensemble SVM algorithms

In this paper, an automatic dialect identification (ADI) system is proposed by extracting spectral and prosodic features for Kannada language. A new dialect dataset is collected from native speakers of Kannada language (A Dravidian language). This dataset includes five distinct dialects of Kannada language representing five geographical regions of Karnataka state. Investigation of the significance of spectral and prosodic variations on five Kannada dialects is carried out. Mel-frequency cepstral coefficients (MFCCs), spectral flux, and entropy are used as representatives of spectral features. Besides, pitch and energy features are extracted as representatives of prosodic parameters for identification of dialects. These raw feature vectors are further processed to get a new derived feature vectors by using statistical processing. In this paper, a single classifier based multi-class support vector machine (SVM) and multiple classifier based ensemble SVM (ESVM) techniques are employed for classification of dialects. The effectiveness and performance evaluation of the explored features are carried out on newly collected Kannada speech corpus, with five Kannada dialects and internationally known standard Intonation Variation in English (IViE) dataset with nine British English dialects. Experimental results have demonstrated that the derived feature vectors performs better when compared to raw feature vectors. However, ESVM technique has demonstrated better performance over a single SVM. Spectral and prosodic features have resulted individually with the dialect recognition performance of 83.12% and 44.52% respectively. Further, the complementary nature of both spectral and prosodic features is evaluated by combining both feature vectors for dialect recognition. However, an increase in dialect recognition performance of about 86.25% is observed. This indicates the existence of complementary dialect specific evidence with spectral and prosodic features. The experiments conducted on standard IViE corpus have shown a higher recognition rate of 91.38% using ESVM. Proposed ADI systems with derived features have shown better performance over the state-of-the-art i-vector feature based systems on both datasets.

Acoustic-phonetic feature based Kannada dialect identification from vowel sounds

In this paper, a dialect identification system is proposed for Kannada language using vowels sounds. Dialectal cues are characterized through acoustic parameters such as formant frequencies (F1–F3), and prosodic features [energy, pitch (F0), and duration]. For this purpose, a vowel dataset is collected from native speakers of Kannada belonging to different dialectal regions. Global features representing frame level global statistics such as mean, minimum, maximum, standard deviation and variance are extracted from vowel sounds. Local features representing temporal dynamic properties from the contour level are derived from the steady-state vowel region. Three decision tree-based ensemble algorithms, namely random forest, extreme random forest (ERF) and extreme gradient boosting algorithms are used for classification. Performance of both global and local features is evaluated individually. Further, the significance of every feature in dialect discrimination is analyzed using single factor-ANOVA (analysis of variances) tests. Global features with ERF ensemble model has shown a better average dialect identification performance of around 76%. Also, the contribution of every feature in dialect identification is verified. The role of duration, energy, pitch, and three formant features is found to be evidential in Kannada dialect classification.

Accent Detection Task to Classify Accented and Non-Accented Speech

This paper presents a brief survey on accent detection, accent identification, and accent classification. Speech processing has becoming more popular and inspiring expanses lately in signal processing area. It is because speech is one of the most natural form of human communication. However, in processing speech signals intrinsically show many variations even without background noise. Two different person can produce different spectrograms when saying the same sentence. Dialect or Accent is one of the most important factors that can influence the Automatic Speech Recognition or ASR performance besides gender (Unsupervised accent class). Many researches show that dialect or accent in speech can significantly affect the speech system performance. Various methods have been used to increase the accuracy of ASR with accent detection, accent identification, and accent classification. Fused i-vector and Phonotactic are the latest technique that shows a significant degree of accuracy. The purpose of this paper is to briefly survey on accent detection, accent identification, and accent classification and discuss the major improvements made in the past almost 10 years of research.

Language model adaptation for language and dialect identification of text

AbstractThis article describes an unsupervised language model (LM) adaptation approach that can be used to enhance the performance of language identification methods. The approach is applied to a current version of the HeLI language identification method, which is now called HeLI 2.0. We describe the HeLI 2.0 method in detail. The resulting system is evaluated using the datasets from the German dialect identification and Indo-Aryan language identification shared tasks of the VarDial workshops 2017 and 2018. The new approach with LM adaptation provides considerably higher F1-scores than the basic HeLI or HeLI 2.0 methods or the other systems which participated in the shared tasks. The results indicate that unsupervised LM adaptation should be considered as an option in all language identification tasks, especially in those where encountering out-of-domain data is likely.