| Cochlear implants (CI) have improved enormously from four decades ago. The development of CI has grown in parallel with advances in technology. CI users benefited from this technological evolution, but as in many other fields of science different obstacles have appeared. CI users commonly complain about the performance degradation when exposed to noisy environments. This problem has been studied in the literature but it has not been until recently that interest and dedication have been increased. Not many solutions have been proposed to improve the performance of CI under noisy environments. In this matter, speech enhancement has been the most important field as it provides different solutions for the underlying problem. Several speech enhancement methods have been broadly studied in the literature including: spectral subtraction, Wiener filtering, and subspace. We introduce in this dissertation new methods for speech enhancement oriented for CI applications to overcome the limitations in noisy environments. Although, in the literature it is commonly assumed noise and speech signals being Gaussian distributed; we focus our attention on a different approach. Our assumption considers the speech envelope signals being Gamma or Laplacian distributed while the noise maintains the same traditional assumption (Gaussian). The noise involved in our study is multi-taker babble, which requires more complex methods to mitigate its effects on speech signals. The proposed algorithms are suitable for currently used strategies in CI applications, but our study focuses primarily on the constant interleaved strategy (CIS) with promising results. Additionally, the proposed algorithms may be embedded in current CI strategies providing strong advantages over conventional methods for speech enhancement in CI applications. |
