India, renowned for its linguistic diversity, faces a significant challenge in cross language communication. With over 1600 spoken languages, the need for effective language translation solutions is pressing. This study explores the potential of speech translation, specifically focusing on two approaches: the cascade method and end-to-end (E2E speech-to-speech translation models. The former entails a sequential technique of automatic speech recognition (ASR) followed by machine translation (MT), allowing customization for accuracy and fluency. While this method has proven successful for English and European languages, its application to Indian languages remains limited. Notably, the incorporation of Long Short-Term Memory (LSTM) algorithms in both ASR and MT systems has shown promise in enhancing translation accuracy within the cascade approach. In contrast, end-to-end speech-to-speech translation models aim to directly map spoken language from one source to a selected target language excluding any explicit linguistic modeling. This approach has shown superior translation accuracy, but requires large computation power and data to train effectively, the latter of which is a significant challenge for the low-resource languages, which most Indian languages are. Ultimately, both methods show promise, and there is wide scope for improvement in both as there has been limited work done in either.