Design and Implementation of a Real-time Transcription Solution

Abstract

In an increasingly interconnected world, effective communication across language barriers is essential. Real-time transcription and translation systems have emerged as solutions to facilitate seamless communication in multilingual settings. This thesis presents a system designed to transcribe English speech, translate it into French, and address challenges in capturing clear audio and handling noisy environments. The system incorporates an automatic speech recognition machine learning model capable of transcribing vocabulary typically used in meetings. It focuses on achieving real-time performance with reasonable latency, even on low-performance hardware. Consequently, our system successfully addressed the challenge of capturing clear audio in noisy environments and transcribing vocabulary commonly used in meetings. Despite acknowledging the constraints in accurately recognizing some words and occasional transcription errors, the system's ability to deliver real-time performance with minimal latency on hardware of modest capabilities is noteworthy and the translation system is robust and Effective at capturing semantics.