MATLAB code for Cochlear Implant Project that aims to explore the nuances of speech-to-speech, melody-to-melody signal processing. Original code written by myself include the code written to replicate Shannon et al's study (cochlear_replicate_study_lowpass, cochlear_replicate_study_hilbert), and the make_band_chimeras_modified code which introduces conflicting information between speech-to-speech/melody-to-melody envelope and fine structures to synthesize chimera sounds. The first part pertaining to the replication of Shannon et al's study involves the reconstruction of the filter bank using the specs described within the study. The filter-bank reconstruction is designed to work with a specified number of frequency bands. In addition, the code also includes the generation of a white noise carrier also referenced in Shannon et al's study.
The second part (multi_band_chimera_low_pass, and make_band_chimeras_modified) involve the exploration of the potential for frequency modulation information to improve cochlear-implant signal processing. In the file "multi_band_chimera_low_pass", the Hilbert transform is done through applying a complex filter to the original signals. In "make_band_chimeras_modified", auditory chimeras were synthesized using a series of speech or melody files placed within the sentences and melodies directories. Further information regarding Chimera signals can be found below. The main purpose of the chimera code is to conduct a simpler version of Smith et al's study regarding envelope and fine structure information within cochlear implant signal processing.
The files "cochlear_replicate_study_lowpass" and "cochlear_replicate_study_hilbert" contain the code used to conduct the study seen in Perceptual Study Results. This replication study was done through conducting a simple sentence recognition experiment where each sentence (.wav file within sentences directory) was played based on the number of frequency bands specified in the "nb" array. Because we had 4 bands, each .wav file was repeated 3 times and sentence recognition averaged for all 20 sentences. The findings of this study support that of Shannon et al's study where the number of noise bands possesses a positive correlation with speech recongnition. The higher the number of noise bands within the filter bank the higher the percentage of speech recognition.