Web-based automated transcription and speaker-label analysis system - ATLAS

Abstract

ATLAS is a web application designed to address the challenges of speech processing, specifically focusing on speech diarization and Automatic Speech Recognition (ASR) services. In many applications, accurately distinguishing between speakers and transcribing their dialogue is crucial, particularly in educational and research contexts. ATLAS leverages NTU Speech Lab’s advanced speech diarization and ASR services, allowing users to either upload recorded audio or conduct live recordings for analysis.

The application not only facilitates the transcription of audio input but also offers an intuitive visualization of speaker labels and their corresponding transcriptions. This thesis documents the comprehensive Software Development Lifecycle (SDLC) of ATLAS, detailing each phase from planning and implementation to deployment. It explores scalability considerations and the performance implications of design choices made throughout the development process. Additionally, the report provides insights on the CI/CD pipeline established to optimize the development and deployment processes.