Deep-learning for conversational speech using semantic textual analysis

• 主要作者: Suthakar, Shiny Gladdys
• 其他作者: Chng Eng Siong
• 格式: Final Year Project
• 語言: English
• 出版: Nanyang Technological University 2022
• 主題: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; Engineering::Computer science and engineering::Computing methodologies::Document and text processing
• 在線閱讀: https://hdl.handle.net/10356/156632

Automatic Speech Recognition (ASR) systems today have a prominent and widespread impact among software applications of different domains. They are usually embedded in the applications to provide user input to the main functionality, hence, acting as the cornerstone of these applications, especially potentially life-saving ones. However, most ASR systems today can only work effectively on formal speech input. They have a lot of room to fully understand speech of colloquial nature. Focusing on English speech in the Singaporean context, this project aims to provide a solution for generating formal semantic equivalents of conversational sentences derived from speech. Thus, acoustic and language models of existing ASR systems can be trained with these mappings from conversational to formal text, thus acquiring better comprehension and performance when receiving informal speech input. Furthermore, this project aims to analyse the semantic similarity performance of deep-learning models in terms of semantically similar formal sentence generation and their use of deep-learning techniques. The experimentation results show that the PEGASUS model performs better holistically. This report will present the proposed solution framework and lay out in detail the components of the project implementation.