Cascaded Speech-to-Speech Translation System Focusing on Low-Resource Indic Languages

This paper work proposes a speech-to-speech translation system to improve translation from Indic languages to English and vice-versa, particularly for those languages with less resources. The major components of the proposed system include Whisper to recognise speech, IndicTrans2 to translate and gTTS to convert the text into speech. The performance of the system was evaluated in terms of BLEU, COMET and WER metrics to analyse the quality and naturalness of the translation output. Finally, the proposed approach efficiently addresses the problems of low-resource languages to translate and provide insights for improving S2S systems in the future