Jean-Marc Valin(Amazon Web Services), Umut Isik(Amazon Web Services), Neerad Phansalkar(Amazon Web Services), Ritwik Giri(Amazon Web Services), Karim Helwani(Amazon Web Services) and Arvindh Krishnaswamy(Amazon Web Services)
Over the past few years, speech enhancement methods based on deep learning have greatly surpassed traditional methods based on spectral subtraction and spectral estimation. Many of these new techniques operate directly in the the short-time Fourier transform (STFT) domain, resulting in a high computational complexity. In this work, we propose PercepNet, an efficient approach that relies on human perception of speech by focusing on the spectral envelope and on the periodicity of the speech. We demonstrate high-quality, real-time enhancement of fullband (48 kHz) speech with less than 5% of a CPU core.