Exploring Wavelet Transform Based Methodology for Cochlear Prosthesis Advanced Speech Processing Strategy

In this paper, a novel strategy for cochlear electrical stimulation based on non-linear human auditory physiological properties was developed. This strategy used the continuous Wavelet Transform (WT) as a speech processing tool. The most distinguished characteristic of the WT tool is its efficient time–frequency resolution. Such advantageous feature for cochlear stimulation could be adaptively adjusted not only by its position parameter that arranges frequency division but also by its scale parameter that could adjust the bandpass width. Following human auditory physiological properties in order to be close to the cochlear behavior, our conceived filter bank structure was divided into 21 frequency bands according to the Equivalent Rectangular Bandwidth (ERB) model. A comparative study was performed between the proposed WT speech processing tool and the classical Short Time Fourier Transform (STFT) speech processing tool. For performance evaluation, some objective speech assessment tests relying on spectral leakage (frequency selectivity) and tracking temporal feature (sensitivity) were performed. In order to evaluate the speech intelligibility, subjective listening tests were also assessed with 50 normal hearing listeners using a specific Cochlear Implant simulator. Experimental results showed that the WT speech processing tool better improves speech understanding compared to the STFT speech processing tool. As for more concretization, and in order to prove the validity of the previously mentioned comparative study, a real-time implementation testing of our strategy performed on a specific Digital Signal Processor (DSP) platform was also described. Using a 720 MHz TMS320C6416 DSP board, the execution of our proposed algorithm took only 10.5 μs per sample. This execution time is sufficient to process the speech signal in real-time.