A New Gate Control Unit-Recurrent Neural Network Structure for Audio-Based Sentiment Analysis

Sunil Thimmaiah1,2,Email

Raghu Jayaramu1

1Department of Electronics & Communication Engineering, The National Institute of Engineering, Mysore, Visvesveraya Technological University, Belagavi 590018, Karnataka, India.
2Department of Electronics & Communication Engineering, Nagarjuna College of Engineering & Technology, Bengaluru, Visvesveraya Technological University, Belagavi 590018, Karnataka, India.

 

Abstract

Sentiment analysis, a crucial task in audio processing, involves the classification of emotions expressed in spoken language. The proposed work is based on a novel Extreme Gradient Boosting (XGBoost)-based structure for emotion-based sentiment analysis from dialect speech samples and the results are compared with traditional techniques to prove its uniqueness in improving the performance. We extract relevant acoustic features from the audio signals, such as Mel-frequency cepstral coefficients (MFCC) coefficients and pitch, and utilize them as input to train an XGBoost classifier. The XGBoost algorithm is an ensemble of decision trees with gradient boosting to learn the sentiment patterns in the audio data. The results demonstrate the effectiveness of the XGBoost model for feature optimization which results in the improvement of the classification of sentiment in audio data. After optimizing the features, the classification of three types of sentiment: positive, negative, and neutral is done by using a novel Recurrent neural network (RNN) structure that incorporates a new Gate Control Unit (GCU) specifically designed for audio-based sentiment analysis, because it has gating mechanisms that regulate the information flow within the RNN, enabling the model to selectively focus on relevant acoustic features and effectively capture sentiment-related patterns in the audio data.

A New Gate Control Unit-Recurrent Neural Network Structure for Audio-Based Sentiment Analysis