A Deep Denoising Sound Coding Strategy for Cochlear Implants

Abstract

Cochlear implants (CIs) have proven to be successful at restoring the sensation of hearing in people who suffer from profound sensorineural hearing loss. CI users generally achieve good speech understanding in quiet acoustic conditions. However, their ability to understand speech degrades drastically when background interfering noise is present. To address this problem, current CI systems are delivered with front-end speech enhancement modules that can aid the listener in noisy environments. However, these only perform well under certain noisy conditions, leaving quite some room for improvement in more challenging circumstances. In this work, we propose replacing the CI sound coding strategy with a deep neural network (DNN) that performs end-to-end speech denoising by taking the raw audio as input and providing a denoised electrodogram, i.e., the electrical stimulation patterns applied to the electrodes across time. We specifically introduce a DNN that emulates a common CI sound coding strategy, the advanced combination encoder (ACE). We refer to the proposed algorithm as 'Deep ACE'. Deep ACE is designed not only to accurately code the acoustic signals in the same way that ACE would but also to automatically remove unwanted interfering noises, without sacrificing processing latency. The model was optimized using a CI-specific loss function and evaluated using objective measures as well as listening tests in CI participants. Results show that, based on objective measures, the proposed model achieved higher scores when compared to the baseline algorithms. Also, the proposed deep learning-based sound coding strategy gave eight CI users the highest speech intelligibility results.

A Fused Deep Denoising Sound Coding Strategy for Bilateral Cochlear Implants

“…Specifically, the independent model struggles to effectively denoise speech when ICRA7 noise is present. This finding contrasts with the outcomes documented in [33]. The discrepancy might be attributed to this study's approach, as highlighted in section II-C, where the SNR calculation is performed at the better SNR side.…”

“…1) Bilateral Deep ACE: This condition closely resembles the baseline scenario (referred to as bilateral ACE) in that it lacks any exchange of information between the listening sides. However, it diverges from clinical ACE sound coding strategies by adopting the newly developed Deep ACE approach, as detailed in [32], [33]. More specifically, Deep ACE substitutes the conventional clinical ACE method with a DNN that takes in raw audio as its input and generates the denoised LGF output p k .…”

“…2) Fused Deep ACE: In this study, we introduce an approach involving integrating two monaural Deep ACE [33] models, with one model associated with each listening side. This integration is achieved through the utilization of fusion layers [36].…”

“…Model optimization was facilitated using the Adam optimizer [38]. The model's training employs a cost function based on mean-squared error (MSE) and binary cross entropy (BCE) for each listening side (for a detailed description of this cost function refer to [33]). The hyperparameter configuration used in this study was slightly modified with respect to the ones shown in [32], specifically the separator module was bigger and the deepenvelope-detector (DED) was also increased in size.…”

“…Recently, drawing inspiration from the Conv-TasNet [23], an end-to-end CI sound coding strategy based on deep learning, termed "Deep ACE." was proposed [32], [33]. This approach was designed to replace the clinically available ACE sound coding strategy, but it could also be used to replace other commercially available ones.…”