RESERVOIR COMPUTING WITH TRUNCATED NORMAL DISTRIBUTION FOR SPEECH EMOTION RECOGNITION

Speech is an effective, quick, and important way for communicating exchanging complex information between humans. Emotions have always been a part of normal human conversation which makes the speech more attractive. Because this major role both emotion, many researchers are inspired by studying Emotion Recognition (SER) still has plenty challenges. In study, we proposed novel reservoir computing approach with initialization random connection weights input weight truncated distribution. Furthermore, Population-Based Training (PBT) adopted to optimize hyperparameters whole Echo State Network (ESN) model significant impact on performance. The bidirectional increase memorization capability, Sparse Random Projection (SRP) was applied dimensional reduction as simple, unsupervised, low complexity approach. speaker-independent strategy employed EMODB SAVEE datasets acted emotion dataset Aibo non-acted dataset. achieved 84.8%, 65.95%, 45.99% unweighted average recalls EMODB, SAVEE, respectively. results show that outperforms recent state-of-the-art studies cheaper computational cost.