Research and Development of Malaysia Court Transcription Continuous Speech Recognition System for Malaysian Malay Language

Lim, Kai Ling (2020) Research and Development of Malaysia Court Transcription Continuous Speech Recognition System for Malaysian Malay Language. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Lim Kai Ling.pdf Restricted to Registered users only Download (1MB)

Abstract

Speech recognition enable a person to control a device without pressing any button, keyboard and mouse. It also able the people to transcribe speech into text for easy reading or reference. It can be applied to various system. For example, the assistance system that lunched by Google, called Google Assistance this system enables a person to navigate in google map, texting, call, paly music, searching just using voice. Furthermore, speech recognition system might make change to the judicial system. Problem such as long case proceeding time and case backlog always exist in the court. Although, Court Recording and Transcription system CRT has implemented to solve these problems but the problem still exist but just slightly reduced. This is due there are only limited days in a year so do limited court working days which also called the court calendar and case rise will have to fit into the busy court time calendar and a court can only listen to that many of case. However, the case proceeding efficiency can be further improve by have a real time speech recognition system. The current system CRT required private transcription vendor to produce the transcription service and require to hire a transcriber for the real time transcript which cause the increase in court proceeding cost and time. In this project, it is to design and develop a real time Malay speech recognition system that somehow able to resolve the problem mentioned or make the court proceeding more efficient, so that it had more time to process more cases. The proposed Malay speech recognition system develop using the framework available in the CMU Sphinx. By using the tool kits provide in the CMU sphinx to design and develop a Malay language court speech recognition system. The system is adapted with Malay speech samples that collected by multiple speaker. The speech used for adaptation is all related to court case proceeding. The approach used to build the project are the Mel-frequency Cepstral Coefficient MFCC feature extraction method and Hidden Markov feature classification/recognition method. The Malay speech recognition system have a graphical user interface that designed by using Microsoft Visual Studio 2019 to convenient the use of end user. Moreover, for speech recognition system that undergone both MLLR and MAP adaptation has the best performance when it tested with test script (speeches that is not adapted). Its average word error rate achieved was 33.537%. average word recognition rate was 77.896% and average sentence error rate was 77.5%.