Deep Denoising for Hearing Aid Applications

Abstract

Reduction of unwanted environmental noises is an important feature of today's hearing aids (HA), which is why noise reduction is nowadays included in almost every commercially available device. The majority of these algorithms, however, is restricted to the reduction of stationary noises.In this work, we propose a denoising approach based on a three hidden layer fully connected deep learning network that aims to predict a Wiener filtering gain with an asymmetric input context, enabling real-time applications with high constraints on signal delay. The approach is employing a hearing instrument-grade filter bank and complies with typical hearing aid demands, such as low latency and on-line processing. It can further be well integrated with other algorithms in an existing HA signal processing chain.We can show on a database of real world noise signals that our algorithm is able to outperform a state of the art baseline approach, both using objective metrics and subject tests.

CLCNET: Deep Learning-Based Noise Reduction for Hearing aids using Complex Linear Coding

“…4. Comparison with WF-approach [1] with the WF approach and shows a significant improvement at −5 dB and 0 dB. Interestingly, limiting the attenuation via ∆SNR t to 14 dB yields slightly better results and smaller interquartile range w.r.t.…”

“…Our framework was implemented in PyTorch [19] using a similar MLP-based architecture (3 fully connected layers with ReLU activations) and temporal context to [1]. Only the input and output layers were modified due to the complex filter bank representation input and coefficient output.…”

“…The scale-dependent SDR did not converge in out experiments. The maximum attenuation of the noise reduction was limited to 14 dB similar to [1], because in practice not all noise should be removed for hearing aid applications. Since the attenuation cannot be limited afterwards, the model was trained to only improve the SNR by 14 dB.…”

“…Advances in deep learning recently improved the performance of noise reduction algorithms [1,2,3]. It is common practice to transform the noisy time-domain signal into a time-frequency representation, for instance using a short-time Fourier transform (STFT).…”

“…It is common practice to transform the noisy time-domain signal into a time-frequency representation, for instance using a short-time Fourier transform (STFT). Usually, only the magnitude of the complex valued spectrogram [1,4] is used for noise reduction. Recent publications though, also focus on incorporating phase information in the reconstructed output by using the phase [5] or the raw complex valued spectrogram [6] as input.…”