CSAVocoder: A Causal Spatial Audio Vocoder Towards Real-Time Spatial Audio Generation

#Zhiyuan Zhu, #Han Wang, et al.

• We introduce CSAVocoder, a strictly causal streaming neural vocoder. It features a Spatial Adaptor to fuse pose information and a Spatial Consistency Discriminator to explicitly supervise inter-channel phase and level differences.
• The model achieves high-fidelity waveform reconstruction while preserving precise spatial rendering, all within a constant memory overhead suitable for real-time streaming.

Comprehensive Benchmarking of Long-Form Speech Generation in Diverse Scenarios

#Changhao Pan, #Rui Yang, #Han Wang, et al.

• We propose LFS-Bench, a standardized benchmark decomposing “long-form quality” into acoustics, semantics, and expressiveness. It includes 1,101 samples spanning 17 diverse scenarios (e.g., dialogues, audiobooks).
• Our extensive experiments reveal that current SOTA models still struggle significantly with consistency and hierarchy in highly expressive scenarios compared to real recordings.

[A Multimodal Evaluation Framework for Spatial Audio Playback Systems: From Localization to Listener Preference]

Changhao Pan, Wenxiang Guo, Yu Zhang, Zhiyuan Zhu, Zhetao Chen, Han Wang, Zhou Zhao ACM MM 2025

[MRSAudio: A Large-Scale Multimodal Recorded Spatial Audio Dataset with Refined Annotations]

Wenxiang Guo, Changhao Pan, Zhiyuan Zhu, Xintong Hu, Yu Zhang, Li Tang, Rui Yang, Han Wang, et al. NeurIPS 2025