This paper introduces a deep learning approach to enhance speech recordings made in a specific environment. A single neural network learns to ameliorate several types of recording artifacts, including noise, reverberation, and non-linear equalization. The method relies on a new perceptual loss function that combines adversarial loss with spectrogram features. Both subjective and objective evaluations show that the proposed approach improves on state-of-the-art baseline methods.

@inproceedings{Su:2019:PM,
   author = "Jiaqi Su and Adam Finkelstein and Zeyu Jin",
   title = "Perceptually-motivated Environment-specific Speech Enhancement",
   booktitle = "ICASSP 2019",
   year = "2019",
   month = may
}