Sixty Years of Frequency-Domain Monaural Speech Enhancement: From Traditional to Deep Learning Methods.
Chengshi Zheng, Huiyong Zhang, Wenzhe Liu and 4 others
PMID 37956661WHAT IT FOUND
Speech noise-reduction methods using deep learning and hybrid approaches usually scored better than traditional methods for simulated normal-hearing listeners, but showed smaller and mixed benefits for simulated hearing-impaired listeners.
Key findings
01For simulated normal-hearing listeners, all deep-learning methods and hybrid methods achieved much better performance than traditional methods, whether input features were compressed or not.
02For simulated hearing-impaired listeners, compressed input features gave lower or comparable speech-quality scores than uncompressed features.
03Deep-learning methods improved simulated hearing-impaired intelligibility scores in most cases below 5 dB SNR, but not at 10 dB SNR.
STILL TO COME
How it was doneWhat they foundWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The paper reports objective scores and simulated listener models, not results from real patients or real hearing aid users. It did not include subjective listening tests with normal-hearing or hearing-impaired listeners, which the authors say are needed to fully evaluate benefits. It focused on single-microphone denoising and did not address competing talkers. Deep-learning methods may have too much delay and computational load for hearing aids; the paper says reducing delay below 4 ms is still challenging. For hearing-impaired listeners, benefits were smaller, and some methods worsened quality scores at high SNRs.
Declared interests
The authors declared no potential conflicts of interest. The work was supported by the National Key Research & Development Program of China (2021YFB3201702).
The easy way to misread this
Do not conclude that these methods are proven to improve real hearing-impaired patients' speech perception. The paper used objective scores and simulated listeners, not clinical outcomes, and for hearing-impaired listeners benefits were smaller and sometimes absent at 10 dB SNR.