NVIDIA 发布 Nemotron-3-Diarization 开源说话人日志模型
原标题：nvidia/Nemotron-3-Diarization
AI 摘要
NVIDIA 在 Hugging Face 发布开源权重说话人日志（diarization）模型 Nemotron-3-Diarization，支持流式与离线推理，最多可区分 8 位说话人。模型采用 Sortformer 架构与 Streaming Sortformer 的 AOSC 缓存和 FIFO 队列机制，单检查点最低输入缓冲延迟 80ms，推荐配置 0.32s，离线配置为 30.4s。模型可用于商业或非商业用途，并提供 NeMo-Speech.cpp 本地 C++ 运行时与 NeMo Speech Python 工具链。
正文节选
--- license: openmdw-1.1 pipeline_tag: voice-activity-detection library_name: nemo tags: - speaker-diarization - streaming-sortformer - speaker-tagging - transformers --- # Nemotron 3 Diarization ## Description **[Nemotron 3 Diarization](https://huggingface.co/blog/nvidia/nemotron-diarization)** is an open-weight speaker diarization model designed to determine "who spoke when" in real-world audio. It supports both streaming and offline inference and handles up to eight speakers. Following S