Deep learning-based speech recognition system for data-scarce language: a feasibility study for a call centre

• Main Author: Muhammad Rais Fawwazi
• Other Authors: Chen Lihui
• Format: Final Year Project
• Language: English
• Published: 2018
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; DRNTU::Engineering::Electrical and electronic engineering::Computer hardware, software and systems
• Online Access: http://hdl.handle.net/10356/76363
• Institution: Nanyang Technological University

To improve customer experience, businesses need a deeper understanding of their customers. Customer service call centres gather personalised information about customers, and hence help us to gain this understanding. However, manual data mining of this information introduces redundancy. This project aims to minimise this redundancy through speech recognition system. This project explores deep learning as a solution to speech recognition. Using Indonesian, we focus primarily on data-scarce language. However, we need to compensate the data scarcity to satisfy the need of abundant data required by deep learning-based models. We focus on transfer learning to help the model learn Indonesian by first learning other languages that have adequate data. Transfer learning improves the model performance significantly, while data collection effort is still required. We also discuss the feasibility of building a speech recognition system within a company for business use. As there are commercial solutions such as Google Cloud Speech API, we might consider them as appealing alternatives. However, by increasing efficiency in the pipeline and create higher-level products based on speech recognition, it might be feasible to build and maintain the system.