Sound field reconstruction using spherical harmonics (SH) has been widely used. However, order-limited summation leads to an inaccurate reconstruction of sound pressure when the reconstructed region is large. The reconstruction performance also degrades when it comes to high frequency. Upscaling ambisonic sound scenes is used to overcome the limitations. In this work, a deep-learning-based method for upscaling is proposed. Specifically, the generative adversarial network (GAN) is introduced. Instead of estimating the SH coefficients, a U-Net-based fully convolutional generator is introduced, which directly outputs the two-dimensional sound pressure. Results show that the proposed method significantly improves the upscaling results compared with the previous deep-learning-based method.
(See document for exact affiliation information.)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
