gms | German Medical Science

Cochlear implant (CI) users struggle to understand speech in noisy conditions. In this work, we present novel source separation architectures to improve speech in noise for CI users. These architectures range from font-end deep neural network algorithms to a novel end-to-end speech coding and denoising sound coding strategy that estimates the electrodograms from the raw audio captured by the microphone. The end-to-end speech coding strategy is a deep neural network that completely substitutes the CI sound coding strategy. We refer to this novel artificial intelligence based sound coding strategy as deepCIS when it substitutes the continuous interleaved sampling (CIS) coding strategy or deepACE when is substitutes the advanced combination encoder (ACE) coding strategy. The deepACE is designed not only to faithfully emulate the coding of acoustic signals that ACE would perform but also to remove unwanted interfering noises, when present, without sacrificing processing latency. The novel end-to-end deepACE 2.0 was optimized using a novel learned envelope detector and a new cost function. DeepACE 2.0 was compared to deepACE 1.0 and to a state of the art TasNet deep network architecture as front-end to the ACE. We will refer to this condition as TasNet+ACE. The models were optimized using CI specific loss functions and evaluated using objective instrumental measures and listening tests in CI participants. In the new study 8 CI users participated in the experiment. Results from objective instrumental measures confirm the results observed in behavioral experiments in CI users. The behavioral experiment consisted of the HSM sentence test in quiet, speech shaped noise and with ICRA7 noise. Speech understanding results in quiet with ACE, deepACE and TasNet+ACE were equivalent. In noise, however, deepACE and TasNet+ACE significantly improved the performance of ACE. The performance with deepACE or with TasNet+Ace was not significantly different in noise, however deepACE reduces the latency of the sound coding strategy. Based on these findings, the present study suggests that Deep ACE 2.0 has the potential as a method to improve speech understanding under noisy conditions for CIs. Still, for the future the number of parameters and complexity needs to be reduced. This work was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) Project ID 446611346.