NVIDIA has released Nemotron 3.5 Transcribe and Granary v2, an open-weight model for multilingual speech processing. Nemotron 3.5 Transcribe can handle transcription, speaker attribution, translation, and audio question answering for up to 40 languages. It achieves top results on the Hugging Face Open ASR Leaderboard and other benchmarks for transcription accuracy and speaker diarization. The accompanying Granary v2 provides a data pipeline and fine-tuning tools to adapt the model for specific domains and accents, demonstrated by a 45.8% reduction in word error rate in a medical case study. AI
IMPACT Enhances multilingual AI capabilities by providing a powerful, adaptable open-source speech processing model.
RANK_REASON Frontier-lab model release with system card and benchmark results. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on Mastodon — mastodon.social →
- Artificial Analysis
- Blackwell
- David Airehrour
- DGX Spark
- Granary v2
- His Girl Friday
- Hugging Face
- ModelScope
- NeMo Speech
- Nemotron 3.5 Transcribe
- NVIDIA
- Nvidia Gpu
- Open ASR Leaderboard
- VoiceArena
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →