Design and Evaluation of Speech Processing Systems for Meetei/Meitei Mayek

The paper focuses on Manipuri, an under-resourced language, and is written in a way that will help eager researchers build a robust speech processing system. To improve the quality of speech processing and language processing systems, the tools and models need to be properly designed and evaluated. Various metrics are considered in the areas of syllabification, speech recognition, speech synthesis, machine translation, etc. Multiple works have demonstrated that authors value accuracy, precision, and scores. Tests are designed to assess data perplexity, naturalness of speech, and intelligibility. Furthermore, the challenges and difficulties associated with the development of various systems are compared. The evaluation metrics, both subjective and objective, are thoroughly discussed, as are various modifications designed to address specific issues arising during data collection, design, and evaluation, mainly in speech recognition and speech synthesis.