Transient Noise Reduction Using a Deep Recurrent Neural Network: Effects on Subjective Speech Intelligibility and Listening Comfort

Abstract

A deep recurrent neural network (RNN) for reducing transient sounds was developed and its effects on subjective speech intelligibility and listening comfort were investigated. The RNN was trained using sentences spoken with different accents and corrupted by transient sounds, using the clean speech as the target. It was tested using sentences spoken by unseen talkers and corrupted by unseen transient sounds. A paired-comparison procedure was used to compare all possible combinations of three conditions for subjective speech intelligibility and listening comfort for two relative levels of the transients. The conditions were: no processing (NP); processing using the RNN; and processing using a multi-channel transient reduction method (MCTR). Ten participants with normal hearing and ten with mild-to-moderate hearing loss participated. For the latter, frequency-dependent linear amplification was applied to all stimuli to compensate for individual audibility losses. For the normal-hearing participants, processing using the RNN was significantly preferred over that for NP for subjective intelligibility and comfort, processing using the RNN was significantly preferred over that for MCTR for subjective intelligibility, and processing using the MCTR was significantly preferred over that for NP for comfort for the higher transient level only. For the hearing-impaired participants, processing using the RNN was significantly preferred over that for NP for both subjective intelligibility and comfort, processing using the RNN was significantly preferred over that for MCTR for comfort, and processing using the MCTR was significantly preferred over that for NP for comfort.

Applications, functions, and accuracy of artificial intelligence in restorative dentistry: A literature review

“…[49][50][51][52][53] ANNs are typically used to process sound, images, and videos. [54][55][56][57][58][59] Recurrent neural networks (RNNs) are a type of NNs where connections between nodes can generate a cycle, that lets the output from some nodes impact the following input to the same nodes. 43 In other words, RNNs have a memory and offer a temporal dynamic behavior.…”

“…Indeed, RNNs are FFNNs that can inherently process variable-length input sequences. 54 A subfield of ML is deep learning (DL) (Figure 1) which is based on ANNs where multiple layers are employed for data processing to obtain progressively higher-level features from data. 60 A DL model can perform complex functions due to having more layers and nodes within a layer.…”

“…Tuning of the weights in a back‐propagation neural network (BPNN) narrows down error rates and creates more accurate and precise models by improving the model's generalization 49–53 . ANNs are typically used to process sound, images, and videos 54–59 . Recurrent neural networks (RNNs) are a type of NNs where connections between nodes can generate a cycle, that lets the output from some nodes impact the following input to the same nodes 43 .…”

“…In other words, RNNs have a memory and offer a temporal dynamic behavior. Indeed, RNNs are FFNNs that can inherently process variable‐length input sequences 54 …”

Evaluation of impulse noise reduction in hearing aids with technical measurements and ratings of discomfort

“…However, other studies show that an increase in discomfort is noticed only within a relative comparison between the aided and unaided condition, and not within a comparison with normal-hearing listeners [13,14]. There are several studies evaluating the benefit of INR, however, the methods and outcomes differ among these studies [15][16][17][18][19]. Most studies report a positive effect of INR when test subjects were asked to rate annoyance, sound quality or loudness comfort.…”