Deep Neural Network Based Noised Asian Speech Enhancement and Its Implementation on a Hearing Aid App

2021 ◽  
Vol 20 (5) ◽  
pp. 1-14
Author(s):  
Xiaoqian Fan ◽  
Bowen Yang ◽  
Wenzhi Chen ◽  
Quanfang Fan
Keyword(s):  
Neural Network ◽  
Speech Enhancement ◽  
Deep Neural Network ◽  
Hearing Aid ◽  
Enhancement Effect ◽  
Frequency Data ◽  
Time Performance ◽  
Testing Data ◽  
The Common ◽  
Speech Clarity

This article studies noised Asian speech enhancement based on the deep neural network (DNN) and its implementation on an app. We use the THCHS-30 speech dataset and the common noise dataset in daily life as training and testing data of the DNN. To stack the frequency data of multiple audio frames to improve the effect of speech enhancement, the system compares the best number of stacked frames during training and testing. At the same time, the influence of training rounds on the PESQ is compared, and the best number of rounds is obtained. On this basis, the best model is implemented on the hearing aid app, and the real-time performance of the device is tested. The experiment shows that based on the DNN, using an appropriate number of rounds for training and using an appropriate number of audio frames stacking to improve the speech enhancement effect, and transplanting this speech enhancement model to the hearing aid app, can effectively improve speech clarity and intelligibility within a reasonable time delay range.

scholarly journals An Improved Fully Convolutional Network Based on Post-Processing with Global Variance Equalization and Noise-Aware Training for Speech Enhancement

2021 ◽  
Vol 25 (1) ◽  
pp. 130-137
Author(s):  
Wenlong Li ◽  
◽  
Kaoru Hirota ◽  
Yaping Dai ◽  
Zhiyang Jia
Keyword(s):  
Neural Network ◽  
Speech Enhancement ◽  
Deep Neural Network ◽  
Voice Conversion ◽  
Post Processing ◽  
Generalization Capability ◽  
Convolutional Network ◽  
Fully Convolutional Network ◽  
Subjective Score ◽  
Model Training

An improved fully convolutional network based on post-processing with global variance (GV) equalization and noise-aware training (PN-FCN) for speech enhancement model is proposed. It aims at reducing the complexity of the speech improvement system, and it solves overly smooth speech signal spectrogram problem and poor generalization capability. The PN-FCN is fed with the noisy speech samples augmented with an estimate of the noise. In this way, the PN-FCN uses additional online noise information to better predict the clean speech. Besides, PN-FCN uses the global variance information, which improve the subjective score in a voice conversion task. Finally, the proposed framework adopts FCN, and the number of parameters is one-seventh of deep neural network (DNN). Results of experiments on the Valentini-Botinhaos dataset demonstrate that the proposed framework achieves improvements in both denoising effect and model training speed.

Machine learning improvements to compressive sensing for speech enhancement in hearing aid applications

2021 ◽  
Vol ahead-of-print (ahead-of-print) ◽  
Author(s):  
Hrishikesh B Vanjari ◽  
Mahesh T Kolte
Keyword(s):  
Neural Network ◽  
Machine Learning ◽  
Hearing Loss ◽  
Compressive Sensing ◽  
Speech Enhancement ◽  
Hearing Aid ◽  
World Health ◽  
Content Type ◽  
L2 Norm ◽  
Reconstructed Signal

Purpose Speech is the primary means of communication for humans. A proper functioning auditory system is needed for accurate cognition of speech. Compressed sensing (CS) is a method for simultaneous compression and sampling of a given signal. It is a novel method increasingly being used in many speech processing applications. The paper aims to use Compressive sensing algorithm for hearing aid applications to reduce surrounding noise. Design/methodology/approach In this work, the authors propose a machine learning algorithm for improving the performance of compressive sensing using a neural network. Findings The proposed solution is able to reduce the signal reconstruction time by about 21.62% and root mean square error of 43% compared to default L2 norm minimization used in CS reconstruction. This work proposes an adaptive neural network–based algorithm to enhance the compressive sensing so that it is able to reconstruct the signal in a comparatively lower time and with minimal distortion to the quality. Research limitations/implications The use of compressive sensing for speech enhancement in a hearing aid is limited due to the delay in the reconstruction of the signal. Practical implications In many digital applications, the acquired raw signals are compressed to achieve smaller size so that it becomes effective for storage and transmission. In this process, even unnecessary signals are acquired and compressed leading to inefficiency. Social implications Hearing loss is the most common sensory deficit in humans today. Worldwide, it is the second leading cause for “Years lived with Disability” the first being depression. A recent study by World health organization estimates nearly 450 million people in the world had been disabled by hearing loss, and the prevalence of hearing impairment in India is around 6.3% (63 million people suffering from significant auditory loss). Originality/value The objective is to reduce the time taken for CS reconstruction with minimal degradation to the reconstructed signal. Also, the solution must be adaptive to different characteristics of the signal and in presence of different types of noises.

Supervised speech enhancement based on deep neural network

2019 ◽  
Vol 37 (4) ◽  
pp. 5187-5201 ◽  
Author(s):  
Nasir Saleem ◽  
Muhammad Irfan Khattak ◽  
Abdul Baser Qazi
Keyword(s):  
Neural Network ◽  
Speech Enhancement ◽  
Deep Neural Network

Deep neural networks–based damage detection using vibration signals of finite element model and real intact state: An evaluation via a lab-scale offshore jacket structure

2020 ◽  
pp. 147592172093261 ◽  
Author(s):  
Zohreh Mousavi ◽  
Sina Varahram ◽  
Mir Mohammad Ettefagh ◽  
Morteza H. Sadeghi ◽  
Seyed Naser Razavi
Keyword(s):  
Neural Network ◽  
Finite Element ◽  
Finite Element Model ◽  
Damage Detection ◽  
Deep Neural Network ◽  
Element Model ◽  
Frequency Data ◽  
Vibration Signals ◽  
Intact State ◽  
The Finite Element Model

Structural health monitoring of mechanical systems is essential to avoid their catastrophic failure. In this article, an effective deep neural network is developed for extracting the damage-sensitive features from frequency data of vibration signals to damage detection of mechanical systems in the presence of the uncertainties such as modeling errors, measurement errors, and environmental noises. For this purpose, the finite element method is used to analyze a mechanical system (finite element model). Then, vibration experiments are carried out on the laboratory-scale model. Vibration signals of real intact system are used to updating the finite element model and minimizing the disparities between the natural frequencies of the finite element model and real system. Some parts of the signals that are not related to the nature of the system are removed using the complete ensemble empirical mode decomposition technique. Frequency domain decomposition method is used to extract frequency data. The proposed deep neural network is trained using frequency data of the finite element model and real intact state and then is tested using frequency data of the real system. The proposed network is designed in two stages, namely, the pre-training classification based on deep auto-encoder and Softmax layer (first stage), and the re-training classification based on backpropagation algorithm for fine tuning of the network (second stage). The proposed method is validated using a lab-scale offshore jacket structure. The results show that the proposed method can learn features from the frequency data and achieve higher accuracy than other comparative methods.

Sign in / Sign up

Export Citation Format

Share Document