Conference Paper 
Open Access
Implementing Two Upmixing Concepts With One Conditional GAN
2026 AES 6th International Automotive Audio Conference · Paper 9 · July 2026
Abstract
Signal processing using artificial intelligence (AI) has gained increasing interest because it outperforms existing solutions in many fields. A significant challenge for deep neural networks lies in meeting strict requirements
regarding latency, computational load and memory, which is vital in automotive audio. This paper presents CUpGAN (Conditional Upmix GAN), a computationally efficient method for extracting upmix signals with low
latency, leveraging signal separation for two upmixing concepts using one conditional generative adversarial network (CGAN). One upmix approach utilize spatial positions of direct sources within the stereo image, allowing
for the distribution of sources around the listener. The second approach separates direct and diffuse signals to create an ambience signal for rear surround loudspeakers. By employing phase-aware loss functions, integrating residual connections in the generator, and training with coherent input and target signals, we achieve high sound quality in the generated signals. The residual signals, i.e., difference between input and generated signals can be used as complementary signals for both upmixing concepts. The proposed method reduces memory as most parameters are shared between both applications, allowing seamless switching between upmixing approaches without the need for parameter loading; instead, a small number of parameters are computed by control network and used to modulate the latent variables. The GAN generator is trained on synthetically generated data, enabling control of separation characteristics that surpass traditional methods. We present an evaluation using listening tests and computational metrics, demonstrating the advantages
of our approach compared to classical signal processing methods.
