Shohojogi: an automated Bengali voice chat system for the banking customer services

Abstract

As the name refers, this research aims to develop an automated voice chat system in Bengali language for the banking system. The system enables clients to learn about detailed information such as account opening queries, loan requirements, fund transfer limits, etc. through a natural language-based interactive voice response system. The system will use speech recognition technology to understand the cus- tomer’s voice commands and respond accordingly in Bangla.It uses the sentence summarization technique and also uses text-to-speech technology to provide spoken responses to the customers. Customer call centers have grown in popularity as a result of pandemics and are now widely employed in a variety of industries, includ- ing e-commerce, hospitals, banks, credit card assistance, and government agencies, among others. Also, it is more difficult to satisfy all of the call center clients due to humans’ constraints on being available 24 hours a day and the variation in waiting times. In order to effectively manage consumers by giving a domain-based answer in the customer’s local tongue, customer service must be automated, especially in emerging nations like Bangladesh where the number of contact support centers is growing. While most people speak in Bangla, there hasn’t been much progress made in automating customer service in the local tongue. By recognizing user voices, defining users’ issues in the standardized Bengali language, and gathering users’ replies into the database to provide feedback in accordance with the queries, our established approach, ”Shohojogi”, can reply to that customer’s requirement. The ability to listen and speak with the user is implemented using speech recognition by the wav2vec2 model while for text summarization we used the seq2seq model and the ability to understand and find the related information is implemented by using the doc2vec model. Finally, we use gTTS for text-to-speech conversion.