A Long Short Term Memory (LSTM) Network based voice sentiment analyzer.
The model can effectively analyze 7 fundamental emotions of Anger, Disgust, Fear, Happiness, Neutrality, Sadness and Surprise.
It was trained on the Toronto Emotional Speech Set (TESS) dataset. This has caused some inherent limitations to the generalizability of the model.
As TESS currently has only two female voice actors and was recorded in a studio setting this model is not at par at being deemed worthy of direct production.
The code is in the jupyternotebook file and is well documented.