Noise enhancement is a critical component of a systems such as Text to speech. A new class of generative model inspied by thermodynamic diffusion have shown primising results on image generation and enhancement. This project takes the generative approach to solve the problem of speech enhancement using waves and shows promising results.
Emotion classification primarily focuses on classification, predicting and retreiving emotion and related properties however only a few researches talk about who is experiencing which emotion, what caused it and what or whom is it directed towards. In this research we try to approach this problem and compare a few models and their performances on this task of semantic role labelling.
Adapting existing OCR systems from Latin to alpha syllabary languages, such as Bengali, is challenging due to sharp contrast in their orthographies. In this project, I try to solve this issue by creating a grapheme detection and segmentation system using a linear pass approach.