Results

Written on April 18th, 2022 by Kelvin Murillo & Shad Fernandez

Emotion Waveform Graphs

neutral1 neutral2

happy1 happy2

sad1 sad2

angry1 angry2

Traditional ML Model Performances

Model Accuracy Precision Recall
Random Forest 71.37% 70.89% 71.58%
Decision Tree 61.74% 62.49% 61.85%
Support Vector Machines 66.42% 67.1% 66.52%

Confusion Matrices

RFConfusionMatrix DecisionTree SVM

Neural Network

Model Accuracy Precision Recall
1D CNN 70.7% 70.68% 70.5%
2D CNN 75.0% 74.3% 74.1%
2D CNN + LSTM 76.6% 76.2% 76.4%

Test Accuracy During Training

ttcnn

ttlstm

Confusion Matrices

CNN1d RFConfusionMatrix RFConfusionMatrix

Speech Emotion Recognition Web Application

Once we gathered all of our results, we built a real-time web application that records a user’s voice and then makes a prediction in real time on their emotion in the recording. Use the following link to try out the Web Application! https://speech-emotion-rec.herokuapp.com/