Bridging Language Barriers: Exploring Hindi-to-English Speech-to-Speech Translation for Multilingual Communication

The increasing demand for cross-lingual communication emphasizes the importance of S2S translation systems, notably in multilingual countries like India. This study develops a structured Hindi-to-English S2S model, achieving a significant MOS of 3.8, bridging language gaps for effective communication. The results can be found at here.

Exploring Current Transformer-Based Models in Speech Processing Tasks: A Concise Review.

Speech processing, a continually evolving field, seeks to improve methods and harness the power of transformer models with their ability to capture both long-term and short-term dependencies effectively. This paper compares Artificial Intelligence techniques, focusing on transformer derivatives like Conformers, which blend convolution and transformers, to provide valuable insights for newcomers and advance the realm of speech processing.

Towards a More Inclusive Telugu Internet: Creating Corpus and Developing Models to Detect Hate Speech in Telugu Code-Mixed Social Media Text

This study centers on detecting hate speech in code-mixed Telugu-English content, a critical task for social media platforms striving to maintain a safe and inclusive online space. Telugu, a language with limited NLP resources, presents challenges due to the scarcity of data. To address this, we've generated a corpus comprising 4500 code-mixed hate speech comments from YouTube. We detail the corpus creation process and share the results of hate speech analysis trained on this dataset.