Enhancing Speech Emotion Recognition based on Feature Level Fusion Using Dual Feature Extraction

ABSTRACT: Automatic speech emotion recognition uses speech clues to figure out what emotions are. Recognising speech emotions (SER) is very important for figuring out what the speaker means. It is important for robots that can feel emotions, interact with people, and social robots. But it is still hard to encode and combine emotional signals. Over the years, researchers have utilised a lot of different classifiers and traits to figure out how people feel when they talk. Voice emotion recognition systems have gotten better, yet they still have problems. For appropriate emotion categorisation, speech data must comprise features that are important to emotions. Handcrafted features were used to sort speech signals by emotion, however they don't work well and don't accurately show how the speaker feels. To get over these limits, we need to improve speech emotion recognition by combining features at the feature level. This study employs Dual Feature Extraction. Bi-LSTM is used by this system to sort speech emotions. GTCC and MFCC are taken out for SER. This study looks at accuracy, sensitivity, precision, and the F1-score.