New IEEE/ACM Journal Article on Robust Features
26 May 2021, by Timo Gerkmann
Our paper "SNR-Based Features and Diverse Training Data for Robust DNN-Based Speech Enhancement" by Robert Rehr and Timo Gerkmann has been accepted for publication in the IEEE/ACM Transactions on Audio, Speech, and Language Processing.
The paper shows how normalizing the input of a neural network using conventional PSD estimators reduces the amount of required training data and increases the generalizability of speech enhancement approaches.
Robert Rehr, Timo Gerkmann, "SNR-Based Features and Diverse Training Data for Robust DNN-Based Speech Enhancement", IEEE/ACM Trans. Audio, Speech, Language Proc., 2021. [doi][arxiv][audio]