Speech Loss Compensation by Generative Adversarial Networks
Yupeng Shi, Nengheng Zheng, Yuyong Kang, Weicong Rong · 2019
Speech loss, including frequency loss and packet loss, can lead to significant speech distortion in many Internet-based speech communication services. In this study, a generative adversarial networks (GANs) structure, which takes deep convolutional neural networks (CNN) as the generator and discriminator components, is adopted as a general framework for speech loss compensation. Network settings are modified for real-time communications. A set of experiments are conducted to evaluate the performance of the GANs-based framework for both bandwidth expansion (BWE) and packet loss concealment (PLC) at several simulated loss conditions. Experimental results demonstrate that the proposed system achieves better performance, with respective to 4 objective metrics, in both BWE and PLC compared to the baseline systems.