Feature optimization for pathological voice based on BP neural network
Weijun Zeng · 2015
The database (Zhao et al. 2013 & Gan et al. 2014) consisted of 151 speakers, 78 healthy speakers and 73 pathological speakers. The voice was recorded in a quiet and small room, where the surrounding noise was below 45 dB. The recordings of the normal quality of voice abnormal quality of voice were obtained from the speakers with disordered speech. For our experiment, the vowel /a/ was sampled from the speakers, and each speaker phonated lasting from 3 to 4 seconds, sampled 3 times, and the sampled frequency was 16 kHz.