Systematic Literature Review and Bibliometric Analysis of Low-Resource Speech-to-Text Translation

This study aims to conduct a comprehensive analysis of scientific literature through bibliometric methods, focusing on the domain of low-resource speech-to-text translation (LRSTT). The primary objective is to investigate the key trends and the annual scientific output in this field. Utilizing bibliometric analysis, authors examine word co-occurrence networks and the distribution of research across different countries. Additionally, we systematically review existing studies on LRSTT to discern prevailing trends within the scientific community. Our methodology involves retrieving scholarly articles from prominent databases such as Web of Science and Scopus, covering the period from 2013 to 2024, and filtering results using PRISMA methodology. From the initial pool of 725 articles, 169 are included in the bibliometric analysis, with a further 15 articles selected for in-depth trend analysis. This paper elucidates critical aspects of LRSTT, including pre-training methods, encoder-decoder architecture, and transformer models, which are pivotal for understanding current advancements in speech-to-text translation.