AIVoiceArena launches Diarization Bench to compare 12 diarization systems
A post describes tests of systems that identify who spoke when, highlighting how timing tolerance around speaker changes affects NVIDIA Nemotron 3's error rate.
TLDR
According to a post about the launch, VoiceArena's Diarization Bench compares 12 systems using the same real-world audio and scoring rules. The post says accuracy drops quickly with overlapping speech and more speakers. It highlights the effect of the “collar”—the timing tolerance around a speaker change: NVIDIA Nemotron 3 scores a 4.29% diarization error rate with a 250 ms collar, but 14.72% when the exact same output is scored with a 0 ms collar.
Combined views
3.6K
2 Sources, first seen 5h ago
VoiceArena launches Diarization Bench to compare 12 diarization systems
A post describes tests of systems that identify who spoke when, highlighting how timing tolerance around speaker changes affects NVIDIA Nemotron 3's error rate.