The acoustic environment is one of the most familiar environments to humans, and interacting with the environment is an important characteristic of embodied intelligence. In the analysis of three-dimensional sound fields, the decomposition of sound into spherical harmonics enables the decoupling of source information and environmental information, making it widely used in three-dimensional sound field analysis based on spherical arrays. In this study, based on the signals received by a spherical array, a spherical harmonic domain expansion is performed. An iterative inversion model training method is employed to couple the acoustic environment with the target signal. The spherical harmonic domain signals are separated in multiple dimensions, including multiple sound sources, early reverberation, late reverberation, and noise. The separated spherical harmonic function signals can be used for tasks such as source analysis and sound field characterization. The related technological achievements have been applied in various fields, including three-dimensional sound field recording and playback, three-dimensional sound field control, source detection, localization, and enhancement, and multi-channel audio coding and decoding.

This research has been supported by the National Key R&D Program, the National High-Tech Development Program (863 Program), the National Natural Science Foundation, and other projects. Multiple research have been published in well-known domestic and international journals and top conferences, including IEEE TASLP, JAES, ICASSP, Interspeech, AAAI, AES Convention, and have been applied for national invention patents. Some patented technologies have been adopted by standards such as 3GPP IVAS and AVS3-P3, and have been implemented in Huawei's MateView, headphones, and other products. The AVS3-P3 codec was used for the 2022 Mid-Autumn Festival Gala and the 2022 FIFA World Cup broadcast.

图：音乐会录制现场