Research On Environmental Acoustic Recognition And Noise Reduction Algorithm Of Cochlear Implant

Cochlear implant(CI) is known as one of the most successful neural prostheses. Its development covers a number of disciplines including medicine, acoustics, electronic engineering, biology, and psychology, etc. Currently most CI users can achieve good speech perception in quiet environments. However, their sound perception abilities, compared to those of normal hearing(NH) people, are still not convincing in noisy conversational environments.This study first carried out a set of psychoacoustical experiments to investigate the performance on environmental sound(ES) recognition by CI recipients. The effects of the vocoder simulation methods(sinusoidal-band vocoder and noise-band vocoder) on ES perception performance were examined via psychoacoustical experiments with 14 NH listeners and 9 CI users as well, all with Mandarin as their native language. The sound materials for the experiments were collected from the Internet and pre-tested by 12 NH listeners to assure the quality of the data. Results showed that the carrier types have no significant effects on the average recognition rates of the 67 vocoded ESs, although they may have some effects on the recognition of some particular vocoded ESs depending on the acoustic characteristics of the original sounds. In addition, the 9 CI users showed poorer performance than NH and the materials in this study can be used in future CI studies.Secondly, we conducted a preliminary study on noise reduction algorithm for CIs to address the problem of poor speech recognition performance in noisy environments. Several Wiener filter based noise reduction algorithms, with an ideal noise estimation and several different parametric Wiener gain functions, were adopted and compared in the study. In particular, three gain thresholds, i.e.,-5, 0 and 5 dB, were set for the Wiener gain functions to explore the effects of gain thresholds on vocoder-simulated-CI performance with different frequency resolutions. The noise reduction performances were evaluated through the vocoder simulation methods(4- and 12-channel sinusoidal-band vocoders) perceptual experiments with 10 NH listeners. Results showed that the Wiener filter based algorithms could provide significant benefits to the simulated-CI hearing, and no significant difference among the three thresholds was found.