NVIDIA 开源 Nemotron 3.5 Transcribe：30B MoE 语音转写模型
NVIDIA Releases Nemotron 3.5 Transcribe: An Open 30B MoE Model for 40-Language Transcription with ~4B Active Parameters
NVIDIA 开源了个语音转写模型，MoE 架构只激活 4.3B 参数，能转 40 种语言还带说话人标签，英文错误率 4.71%，可以自己部署。
NVIDIA 发布开源权重语音模型 Nemotron 3.5 Transcribe，总参数 30B 的 MoE 架构，每次推理仅激活约 4.3B 参数。该模型支持 40 种语言的语音转写，可输出说话人标签和词级时间戳。在英文 Open ASR Leaderboard 上，它的平均词错误率（WER）为 4.71%。
NVIDIA Releases Nemotron 3.5 Transcribe: An Open 30B MoE Model for 40-Language Transcription with ~4B Active Parameters
NVIDIA released Nemotron 3.5 Transcribe, an open-weights speech LLM with 4.3B active parameters. It transcribes 40 languages, adds speaker labels and word timestamps, and posts a 4.71% mean WER on the English Open ASR Leaderboard. The post NVIDIA Releases Nemotron 3.5 Transcribe: An Open 30B MoE Model for 40-Language Transcription with ~4B Active Parameters appeared first on MarkTechPost .
