A deep learning approaches in text-to-speech system: a systematic review and recent research perspective

Abstract

Text-to-speech systems (TTS) have come a long way in the last decade and are now a popular research topic for creating various human-computer interaction systems. Although, a range of speech synthesis models for various languages with several motive applications is available based on domain requirements. However, recent developments in speech synthesis have primarily attributed to deep learning-based techniques that have improved a variety of application scenarios, including intelligent speech interaction, chatbots, and conversational artificial intelligence (AI). Text-to-speech systems are discussed in this survey article as an active topic of study that has achieved significant progress in the recent decade, particularly for Indian and non-Indian languages. Furthermore, the study also covers the lifecycle of text-to-speech systems as well as developed platforms in it. We performed an efficient search for published survey articles up to May 2021 in the web of science, PubMed, Scopus, EBSCO(Elton B. Stephens CO (company)) and Google Scholar for Text-to-speech Systems (TTS) in various languages based on different approaches. This survey article offers a study of the contributions made by various researchers in Indian and non-Indian language text-to-speech systems and the techniques used to implement it with associated challenges in designing TTS systems. The work also compared different language text-to-speech systems based on the quality metrics such as recognition rate, accuracy, TTS score, precision, recall, and F1-score. Further, the study summarizes existing ideas and their shortcomings, emphasizing the scope of future research in Indian and non-Indian languages TTS, which may assist beginners in designing robust TTS systems.