APPLIED LINGUISTICS WITH DEEP LEARNING-BASED DATA-DRIVEN TEXT-TO-SPEECH SYNTHESIZER FOR ARABIC CORPUS

Abstract

The field of Applied Linguistics, which deals with language and its practical uses, connects with technology in interesting methods, especially in the advancement of text-to-speech (TTS) synthesizers. TTS synthesizers change written text into spoken words, deploying ethics from phonology, phonetics, and syntax to create natural-sounding speech. Within linguistics use, these methods are invaluable purposes such as increasing communication in fractal human–computer interactions (HCIs), language-learning tools, and providing accessibility solutions for visually impaired individuals. TTS purposes at synthesizing understandable and natural speech from text, and it has advanced quickly in recent times because of the progress of artificial intelligence (AI). During past years, deep learning (DL)-based TTS techniques have been established quickly, enabling the generation of natural speech with a high-quality narrator that matches human levels. Creating TTS methods at the quality of the human level has always been the aspiration of speech synthesis practitioners. Although current TTS techniques achieve impressive voice quality, there remains an evident gap in quality compared to human recordings. In this paper, we present an Applied Linguistics with Deep Learning-based Data-Driven Text-to-Speech Synthesizer (ALDL-DDTTS) technique for Arabic corpus. The ALDL-DDTTS technique mainly aims to detect the text and convert it into speech signals on Arabic corpora. In the ALDL-DDTTS technique, a multi-head attention bi-directional long short-term memory (MHA-BiLSTM) approach can be employed with fractal optimization methods to predict the diacritic and gemination signs. Additionally, the Buckwalter code has been deployed for capturing, storing, and displaying the Arabic texts. To boost the efficiency of the ALDL-DDTTS technique, the hyperparameter selection process uses the fractal ant lion optimization (ALO) algorithm. For examining the boost performance of the ALDL-DDTTS methodology, a wide range of simulations is involved. The experimental outcomes illustrated that the ALDL-DDTTS technique reaches better performance than other models.