Voice Technology and AI: Innovative Tools for Developing Inclusive Education, Disability Learning, and Mental Health Support

Abstract

This dissertation tackles a major obstacle in Natural Language Processing (NLP)
related to the analysis of unstructured and noisy utterances, specifically in Indian regional
languages. The training of NLP algorithms on corpora primarily available in languages like
English, French, and Chinese has limited applicability to regional languages due to the lack
of relevant corpora. This issue becomes more pronounced in real-world scenarios where
the speech is often distorted by background noise and non-sophisticated Automatic Speech
Recognition (ASR) engines.
To address this problem, the dissertation proposes an innovative solution that uses
ensemble methods and focuses on the accuracy of uttered sentences. The approach
combines rule-based Machine Learning and fuzzy logic-based algorithms to overcome
language variances and accurately identify full or partial matches of actual answers within
noisy utterances.

The methodology is adapted based on the language of the uttered answers. For
regional languages, the dissertation employs sequence matching and fuzzy matching
algorithms, with special consideration for answers containing numbers through the
utilization of number-to-word methods. In the case of English answers, the methodology
involves word form, fuzzy logic, and distance-based similarity matching.
The detailed description of the proposed solution emphasizes the inventive aspects
contributing to its efficacy. While specific diagrams or figures illustrating the components
are not provided, the absence of patented algorithms in the literature or prior art section
underscores the novelty of the proposed solution, presenting a distinct and valuable
contribution to NLP for regional languages.
In summary, this dissertation presents a groundbreaking approach to address the
challenges posed by unstructured and noisy utterances in NLP, particularly in Indian
regional languages. The ensemble of algorithms and language-specific methodologies
positions it as a novel and promising solution capable of significantly enhancing the
accuracy of understanding and matching uttered sentences.