Use of a Deep Recurrent Neural Network to Reduce Wind Noise: Effects on Judged Speech Intelligibility and Sound Quality

Abstract

Despite great advances in hearing-aid technology, users still experience problems with noise in windy environments. The potential benefits of using a deep recurrent neural network (RNN) for reducing wind noise were assessed. The RNN was trained using recordings of the output of the two microphones of a behind-the-ear hearing aid in response to male and female speech at various azimuths in the presence of noise produced by wind from various azimuths with a velocity of 3 m/s, using the “clean” speech as a reference. A paired-comparison procedure was used to compare all possible combinations of three conditions for subjective intelligibility and for sound quality or comfort. The conditions were unprocessed noisy speech, noisy speech processed using the RNN, and noisy speech that was high-pass filtered (which also reduced wind noise). Eighteen native English-speaking participants were tested, nine with normal hearing and nine with mild-to-moderate hearing impairment. Frequency-dependent linear amplification was provided for the latter. Processing using the RNN was significantly preferred over no processing by both subject groups for both subjective intelligibility and sound quality, although the magnitude of the preferences was small. High-pass filtering (HPF) was not significantly preferred over no processing. Although RNN was significantly preferred over HPF only for sound quality for the hearing-impaired participants, for the results as a whole, there was a preference for RNN over HPF. Overall, the results suggest that reduction of wind noise using an RNN is possible and might have beneficial effects when used in hearing aids.

Using recurrent neural networks to improve the perception of speech in non-stationary noise by people with cochlear implants

“…The LSTM processed a 5-timestep input where each timestep was related to acoustic features extracted from a single frame of the input signal (noisy speech); steps 1, 2, 3, 4, and 5 corresponded to successive frames j -4, j -3, j -2, j -1, and j , respectively. We selected this architecture based on previous studies using HI listeners (Keshavarzi et al ., 2018; 2019). The RNN estimated the IRM for frame j as its output (estimated ratio mask, ERM).…”

“…RNN-LSTM algorithms have shown improved generalization using objective measures, but have not been evaluated in listening studies with CI users. However, similar types of LSTM-RNNs have recently been shown to provide benefits for speech-in-noise perception for HI listeners (Bramslow et al ., 2018; Keshavarzi et al ., 2018; 2019; Healy et al ., 2019), and they represent a promising way for improving performance for CI users in conditions with non-stationary noise that was not included in the training data.…”

Using recurrent neural networks to improve the perception of speech in non-stationary noise by people with cochlear implants

“…The LSTM processed a five-timestep input wherein each timestep was related to acoustic features extracted from a single frame of the input signal (noisy speech); steps 1, 2, 3, 4, and 5 corresponded to successive frames j-4, j-3, j-2, j-1, and j, respectively. We selected this architecture based on previous studies using HI listeners (Keshavarzi et al, 2018;Keshavarzi et al, 2019). The RNN estimated the IRM for frame j as its output (estimated ratio mask, ERM).…”

“…RNN-LSTM algorithms have shown improved generalization using objective measures, but have not been evaluated in listening studies with CI users. However, similar types of LSTM-RNNs have recently been shown to provide benefits for speech-in-noise perception for HI listeners (Bramsløw et al, 2018;Keshavarzi et al, 2018;Keshavarzi et al, 2019;Healy et al, 2019), and they represent a promising way for improving performance for CI users in conditions with non-stationary noise that was not included in the training data.…”

PM2.5 concentration forecasting using Long Short-Term Memory Neural Network and Multi-Level Additive Model

“…Statistical methods can predict the concentration of air pollutants including PM 2.5 by analyzing air quality related data and have received extensive attention from scholars. [15] models, autoregressive moving average (ARIMA) [16] models, land use regression (LUR) [17] models, generalized additive 4 models (GAM) [18,19], support vector regression (SVR) models [20], artificial neural network (ANN) models [21] in machine learning, recurrent neural network (RNN) [22,23], convolutional neural networks (CNN) [24,25] and long-term memory neural network (LSTM) [26,27] in deep learning [28].…”