Speech emotion recognition in noisy environment

Under real life condition, speech signal is often, corrupted with several noise types. To attenuate this issue, a noise reduction phase is performed before analyzing emotional speech using enhancement algorithms. Three speech enhancement algorithms are introduced for improved emotion classification; spectral subtraction, wiener filter and MMSE. Experiments were prepared with MFCC as feature vectors and HMM as classifier. Experiments are evaluated on real condition speech signal (IEMOCAP database) with real world noise using various SNR level. Results after denoising were compared to those before denoising and those without noise to measure the system performance. The experimental results show that the speech enhancement algorithms improve the performance of our emotion recognition system under various SNRs.