Emotion Detection through Speech using Bidirectional LSTM and Attention Mechanism

  • Megha Agarwal et al.


Artificial Intelligence is growing and developing at an exceptional rate. Artificial machines and robots
are incorporated with various ways to handle different scenarios and come up with accurate solutions
through artificial intelligence. However, when it comes to taking some decisions based on emotions
and including emotional quotient in the decision-making process, artificial machines face some issues.
Apart from this, embedding emotions into Artificial intelligence just widens the scope for various
further researches. To work on improving the emotional aspect in artificial intelligence systems, we
need to first tackle the issue of detecting emotions with least possible errors. In this paper the aim is to
find ways to improve upon accuracy in emotion detection through deep learning. Deep learning
methods work by processing a vast database gathered from a number of sources. The analysis initiates
by vectorizing each word in the input given by the user and deriving the meaning of the words in both,
forward and backward direction. Upon understanding the meaning, attention mechanism defines the
weights to be assigned to the words based on the importance they carry. This results in a maximum
pooling of the highest weight vectors. The vectors then proceed to be classified in one of the six major