Autonomous language processing for business solutions

• Main Author: Kor, Jia Li
• Format: Final Year Project / Dissertation / Thesis
• Published: 2021
• Subjects: HG Finance
• Online Access: http://eprints.utar.edu.my/5217/1/1705353_FYP.pdf; http://eprints.utar.edu.my/5217/

A speech analytics solution worked with the combination of speech recognition and Natural Language Processing (NLP). It converted spoken sentences into written words by using Python Programming and with the help of Google Cloud Speech To Text API. Speech recognition steps included receiving "speech" either through microphone or audio files firstly. Then, the “speech” converted from physical sound into an electrical signal. The electrical signal was then being converted into digital data using an analogue-to-digital converter. Lastly, a model was used to convert the audio into text once it has been digitized. NLP helped a computer to understand languages spoken by humans. It was explained as an automated way of analysing the written text by following some theories and technologies. In the business area, speech analytics were used to make predictions and developed an understanding of the clients’ metrics . In this study, we focused on the languages such as Malay languages and mixed languages which were commonly used in Malaysia. Most of the call recordings data that used were basically containing these two languages. As Malay and mixed languages were not the worldwide languages, it increased the difficulty of developing a speech analytics solution that converted these two languages accurately into written text. Therefore, we expected that the results of this research improved the accuracy of speech analytics solutions so that it increased the efficiency of the insurance company in dealing with their clients. The accuracy of the speech analytics solutions in converting the spoken word into written text was investigated with Word Recognition Rate and an accuracy scale table used as a reference. There were two factors such as “Time Cut Point” and audio’s speed, being investigated in order to determine whether it would bring any effect towards the accuracy of text transcription. Different “Time Cut Point” and audio’s speed used in manipulating the data. Both factors were analysed together in a combination form. The best combination was chosen for both evaluation methods (WRR and accuracy scale table).