Stacked Convolutional Autoencoder with Multi-label Extreme Learning Machine (SCAE-MLELM) for Bangla Regional Language Classification.

Several studies emphasized the importance of detecting the dialect spoken by the user, to improve customer-agent relationships in call centers operated by telecomunication companies. and to create advanced speech recognizing Artificial Intelligence devices using machine learning (ML) techniques. However, the extensive Bangla speech recognition research performed in the past did not address this issue. Therefore, in this paper, we propose a model that uses deep learning (DL) techniques, as a subset of ML, to classify the regional language in the audio signal of the Bangla speech. We build a model based on Stacked Convolutional Autoencoders (SCAE) and Multi-label Extreme Learning Machines (MLELM). SCAE extracts the spatial and temporal features from the input data and produces a detailed feature vector. In order to provide a soft classification score and hard labels, the information is passed onto a series of MLELM networks. For Bangla speech dataset created during this research, the proposed method produced an accuracy score of 95%. Furthermore, the proposed system performs remarkably well in classifying the age and gender of the speaker in Bangla.