公开(公告)号：WO2020139588A1

公开(公告)日：2020-07-02

申请号：PCT/US2019/066315

申请日：2019-12-13

Applicant: DTS, INC.

Inventor： WALSH, Martin ,  MCDONAGH, Aoife ,  GOODWIN, Michael, M. ,  STEIN, Edward ,  CORCORAN, Peter

IPC: H04S7/00 ,  G06F30/20 ,  G06K9/00 ,  G06T17/20

Abstract: A method comprises: receiving an image of a real-world environment; using a machine learning classifier, classifying the image to produce classifications associated with acoustic presets for an acoustic environment simulation, the acoustic presets each including acoustic parameters that represent sound reverberation; and selecting an acoustic preset among the acoustic presets based on the classifications.

公开(公告)号：WO2020010338A1

公开(公告)日：2020-01-09

申请号：PCT/US2019/040739

申请日：2019-07-05

Applicant: DTS, INC.

Inventor： LEMLEY, Joseph ,  MCDONAGH, Aoife ,  CASSIDY, Ryan, J.

IPC: G10L25/30 ,  G06N3/04

Abstract: A hybrid audio synthesis system and method for synthesizing hybrid audio from at least two distinct classes of sounds. Embodiments of the system and method include generating a synthesized hybrid audio clip using a generative adversarial network (GAN) that has been trained on two distinct classes of sounds. Embodiments of the system and method also include techniques for generating an extended-duration synthesized hybrid audio clip. This includes techniques for combining synthesized hybrid audio clips. In some embodiments a long short-term memory (LSTM) used to learn the best next audio input vector that is input to the trained GAN generator. In some embodiments the extended-duration synthesized hybrid audio clip is obtained using a Fourier transform of the synthesized hybrid audio clip (such as the Griffin-Lim algorithm or any variation thereof). In some embodiments cross fading is used to seamlessly stitch together the audio segments or clips.