Effects of channel carriers (vocoders) on Chinese speech recognition in continuous interleaved sampling (CIS) and advanced combination encoding (ACE) simulation of cochlear implant

In speech recognition, tone or noise vocoder used as channel carrier has been used in experiments to simulate cochlear implant processing in normal hearing (NH) listeners. The purpose of this research is to study what is the difference between the white noise and sine wave vocoders on Chinese speech recognition in cochlear implant (CI) simulation. Two encoding strategies (CIS and ACE) were simulated to evaluate Chinese word and tone recognition rates on eight male subjects (age 23-26) with normal hearing in our experiments. Single character Chinese words drawn from the open-set of pre-school children Mandarin Monosyllabic Lexical Neighborhood Test were used as the stimuli for the experiments. These stimuli (100 words organized as four 25-word lists) were recorded by one female speaker with an F0 ranging from 100 to 200 Hz (same difficulty with 25 words/list, 16-bits, sampling rate of 16kHz.). Different channel numbers (4, 8, 12, 16 channels) were tested to evaluate Chinese word and tone recognition rates for CIS and ACE encoding strategies on both vocoders. In addition, different stimulation rates (250Hz, 900Hz, 1200Hz, 1800Hz) with ACE strategy were used to evaluate Chinese word and tone recognition rates for both vocoders. Preliminary results on Chinese speech recognition rates show that there is a statistically significant difference with increasing channel number in CIS strategy on both vocoders, but only statistically significant difference between 4 and 8,12,16 channels in ACE strategy on both vocoders. In addition, there is a statistically significant difference between 250 Hz and 900Hz, 1200Hz, 1800Hz in ACE strategy for Chinese words recognition rates on white noise vocoder, but not sine wave vocoder. Our preliminary results suggest that it is appropriate to use white noise vocoder in Chinese CI simulation.