Depression detection using the derivative features of group delay and Delta phase spectrum

Using speech signal processing technology to detect depression arouses more and more interest recently. But even so, the source of feature extraction is only restricted in the amplitude representation and the important role of the phase, the other half part of signal information, has been ignored. It is the first time that this paper pays attention to the effectiveness of phase spectrum-based feature derived from group delay and Delta phase spectrum for depression detection. The experimental results demonstrate that Mel-Frequency Delta-phase(MFDP) can achieve the modest improvement over Mel-Frequency Cepstral Coefficients (MFCC) in the female group testing with the accuracy of 82.94% and gender-independent tasks with the accuracy of 76.33%. In most cases, MFDP as the derivative feature of Delta phase spectrum is superior to those of group delay.