Automatic Speech Recognition and Translation for Low Resource Languages

5Domain Adaptation-Based Self-Supervised ASR Models for Low-Resource Target Domain

L. Ashok Kumar¹*, D. Karthika Renuka¹, Naveena K. S.² and Sree Resmi S.²

¹PSG College of Technology, Coimbatore, India

²UG Students, Dept. of IT, PSG College of Technology, Coimbatore, India

Abstract

Domain adaptation is the concept of improving the performance of a model on a target domain, by leveraging the knowledge gained from the source domain. Speech recognition, or speech-to-text, is the ability of a machine or a system or a program to identify words spoken by a person and convert them into readable text. In this chapter, a novel concept called the domain adaptation was used on the well-trained, self-supervised automatic speech recognition (ASR) models for the low-resource target domain data. Research in computer science, linguistics, and computer engineering are all used in speech recognition. Speech recognition features are built into a lot of contemporary gadgets and text-focused software to facilitate easier or hands-free usage. Automatic speech recognition is the process of using computer algorithms to convert spoken language into written text. Some of the challenging tasks are being able to recognize speech in noisy environments, the accent difference since it changes from one person to another, and the availability of the dataset to train the model. These models perform well on data which are the same type of data or same data used to train the model but its performance decreases ...