Emotional speech recognition based on ECG

Application of Principal Component Analysis (PCA) for emotional speech recognition using Electrocardiogram (ECG) parameters and prosodic parameters is presented in this paper. We choose people who speak stand mandarin to record emotional speeches and measure the ECG during recording. The ECG signal is filtered using wavelet transform to remove power line interference, base line wander and electromyography interference. R wave amplitude, RR interval and QRS complex duration is selected as ECG parameters. Combined with prosodic parameters, such as amplitude energy, fundamental frequency, first formant´s frequency and so on, the emotional speech can be converted into a ten-dimension eigenvector. Using PCA algorithm, the emotional recognition experiment is carried out, and the result show that the recognition rate based on ECG parameters and prosodic parameters is obviously better than that based on prosodic parameters with 3 to 4 percentage points higher on average.