Automated Recognition, Processing and Sentiment Analysis of Speech

• Designed an LSTM-based recurrent neural network for speech recognition using Mel Frequency Cepstral Coefficients (MFCC) and compared its performance to the baseline Convolution Neural Network model using confusion matrix
• Evaluated and then fine-tuned a pre-trained CRDNN model and utilized it to design a sentiment detection model
• Obtained word error rate of 0.37 and sentiment detection accuracy of 76% for our model on the Hugging Face dataset

Download Presentation ; Download Poster ; Github Repository