Novel Ensemble Sentiment Classification through Speech Processing and Stacking Generalization

This paper introduces an innovative method for speech sentiment analysis, by employing a stacking classifier. The proposed system directly transcribes audio and extracts essential features for sentiment analysis. The stacking classifier, which combines multiple classifiers, enhances predictive performance. Intermediate models include K-Nearest Neighbors (k-NN), Random Forest, SVM, Gaussian Naive Bayes, and Logistic Regression. OpenAI Whisper transformer is used for audio-to-text conversion. The methodology encompasses audio data preprocessing, text conversion, natural language processing, and model training. The model is evaluated on accuracy, recall, precision, and F1-score. We examine the effectiveness of various classifiers and feature sets in sentiment analysis of speech data through empirical investigation. The experimental results demonstrate the system’s ability to accurately classify sentiment and provide valuable insights into sentiment analysis methods, tailored for spoken language.

in 2024 First International Conference on Electronics, Communication and Signal Processing (ICECSP)