Abstract

Speech recognition is a technique employed to convert spoken words of customers into text, which can then be analyzed through various analytical models to gauge customer satisfaction. However, solely analyzing the words without considering the tone of speech may not accurately represent customer responses. Customer satisfaction is particularly crucial for shopping centers as it directly impacts their growth.

Prior research has addressed customer satisfaction confirmation through speech analysis, which has propelled shopping centers to the next level of growth. This confirmation is collected by recording customer speeches in the Mandarin language. Encoders are utilized to reduce input data dimensionality, while decoders reconstruct the input data. Long-Short Term Memory (LSTM) and Recurrent Neural Network (RNN) models help manipulate the data series, and Support Vector Machine (SVM) models process data classification. Mel Frequency Cepstral Coefficient (MFCC) is employed to classify voice data but is limited to processing frequencies below 1000Hz.

This proposed work aims to enhance the performance of the classification model. The quality of recorded customer speeches and surrounding noise can introduce variations in customer speech. To mitigate environmental noise, Audacity 2.3 is used. The customer speech is recorded in the English language, and LSTM, RNN, MFCC, and SVM models are employed to predict customer satisfaction. By leveraging these techniques, this study seeks to provide valuable insights into the customer experience and satisfaction levels, contributing to the growth and success of shopping centers.

Keywords: Speech recognition, Customer satisfaction, Analytical models, Survey, Tone of speech, Shopping centers, Mandarin language, Encoders, Decoders, Long-Short Term Memory (LSTM), Recurrent Neural Network (RNN), Support Vector Machine (SVM), Mel Frequency Cepstral Coefficient (MFCC), Environmental noise.