A Multi-Task Scheme for Supervised DNN-Based Single-Channel Speech Enhancement by Using Speech Presence Probability as the Secondary Training Target

To cope with complicated interference scenarios in realistic acoustic environment, supervised deep neural networks (DNNs) are investigated to estimate different user-defined targets. Such techniques can be broadly categorized into magnitude estimation and time-frequency mask techniques. Further, the such as Wiener gain estimated directly or derived by power spectral density (PSD) signal-to-interference ratio (SIR). In this paper, we propose incorporate multi-task learning DNN-based single-channel speech enhancement using presence probability (SPP) a secondary target assist main task. The domain-specific information is shared between two tasks learn more generalizable representation. Since performance of network sensitive weight parameters loss function, homoscedastic uncertainty introduced adaptively weights, which proven outperform fixed weighting method. Simulation results show proposed scheme improves overall compared conventional single-task methods. And joint direct SPP yields best among all considered