Non-spectral features-based robust speaker independent emotion recognition from speech signal