Sentiment Analysis from Speech Signals using Convolution Neural Network

Sentiment analysis for emotion recognition from the speech is the most effective method for interaction of human with machines. It has obtained adequate popularity in present days with implementations in social media, medical field, traffic, customer review, lie detection, carboard system and many more. Numerous methods such as artificial neural network (ANN), recurrent neural network (RNN), and convolution neural network (CNN) are suggested to recognize sentiments from speech. In this paper, we introduce a model with using 1-dimensional CNN consisting of 7 sets of 1D convolution layers, 3 fully connected layers, and an output layer. Acoustic features are extracted from the audio files using different feature extraction technique. The paper considers wave plot as well as spectrogram related features. For increasing data points, data augmentation technique is used, which has helped to improve the classification accuracy. The experimental results validates that the proposed model has performed better in comparison to the existing methodologies.