Kurdish Dialect Recognition using 1D CNN

Dialect recognition is one of the most attentive topics in the speech analysis area. Machine learning algorithms have been widely used to identify dialects. In this paper, a model that based on three different 1D Convolutional Neural Network (CNN) structures is developed for Kurdish dialect recognition. This model is evaluated, and CNN structures are compared to each other. The result shows that the proposed model has outperformed the state of the art. The model is evaluated on the experimental data that have been collected by the staff of department of computer science at the University of Halabja. Three dialects are involved in the dataset as the Kurdish language consists of three major dialects, namely Northern Kurdish (Badini variant), Central Kurdish (Sorani variant), and Hawrami. The advantage of the CNN model is not required to concern handcraft as the CNN model is featureless. According to the results, the 1 D CNN method can make predictions with an average accuracy of 95.53% on the Kurdish dialect classification. In this study, a new method is proposed to interpret the closeness of the Kurdish dialects by using a confusion matrix and a non-metric multi-dimensional visualization technique. The outcome demonstrates that it is straightforward to cluster given Kurdish dialects and linearly isolated from the neighboring dialects.

Klasifikasi Dialek Pengujar Bahasa Inggris Menggunakan Random Forest

Speech recognition is one of the important research fields which is currently widely used for various applications. However, speech recognition performance is affected by the dialect of the speaker. Therefore, dialect recognition is often used as an additional feature in speech recognition. The process of recognizing dialects is not easy. Currently, Machine Learning technology is widely applied in dialect recognition. One of the challenges in the introduction of machine learning-based dialects is the imbalance of classes and overlaps in a wide variety of classification techniques. This study applies Random Forest-based oversampling technology for dialect recognition. For hyper-parameter optimization of the random forest algorithm, we apply the Grid Search method. Experiments on Speech Accent Archive data using the MFCC feature resulted in an accuracy of 0.91 and AUC of 0.95

Sounding like a father: The influence of regional dialect on perceptions of masculinity and fatherhood

Previous work on the Osaka dialect (OD) collectively suggests that this western regional variant of Japanese is associated with informality, masculinity, and affective fatherhood—social meanings that can be recruited in the construction of audio-visual media personas. This study examines the use of OD by one protagonist in the film Soshite chichi ni naru/Like father, like son, as well as the social meanings that listeners attribute to this variety of Japanese. Specifically, we ask two questions: (i) to what extent is the production of OD in the film recognizable to native speakers of Japanese, and (ii) what qualities do Japanese language users attribute to OD? A dialect recognition experiment found low recognizability of OD but high recognizability of a general ‘nonstandard Japanese’ language variety. Qualitative data revealed that Japanese language users perceived OD to index various characteristics including that of a masculine, affective father. (Perception, dialect, fatherhood, Osaka dialect, indexicality)*

Spoken Arabic dialect recognition using X-vectors

This 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.