high performance computing on graphics processing units: hgpu.org

This article presents a low-latency, online speaker diarization system ("who is speaking now?") based on the repeated execution of a GPU-optimized, highly efficient offline diarization system ("who spoke when"). The system fulfills all requirements of the diarization task, i.e., it does not require any a priori information about the input, including specific speaker models. In contrast to earlier attempts at online diarization, the system achieves similar accuracy to the underlying offline system and does not require explicit detection of new speakers. Using GPUs, online diarization has become a side-effect of offline diarization, obsoleting the requirement for specialized online diarization systems.