Gemini 4 Carbon: Our next frontier model
deepmind.google·5d ago
AI Engineering
Hugging Face Blog·5d ago·5 min read
NVIDIA Nemotron 3.5 Transcribe and Granary v2: Intelligent Multilingual Transcription You Can Prompt and Adapt
TL;DR
NVIDIA Nemotron 3.5 Transcribe is an advanced multilingual transcription model that integrates transcription, speaker attribution, translation, and audio Q&A, achieving top leaderboard results. It supports 40 languages and adapts to domain-specific vocabulary, making it ideal for diverse applications.
✦ Why It Matters
Explore NVIDIA Nemotron 3.5 Transcribe today to enhance your multilingual transcription capabilities in applications.
Key Takeaways
- Utilize Nemotron 3.5 for accurate multilingual transcription.
- Leverage Granary v2 for fine-tuning to specific vocabularies.
- Implement prompt-based outputs for tailored transcription results.
- Explore hosted endpoints for easy deployment without self-hosting.
Full Summary
NVIDIA Nemotron 3.5 Transcribe is an open-weight model designed for multilingual transcription, speaker attribution (diarization), translation, and audio question answering. It supports 40 languages for transcription, translates 31 languages to or from English, and attributes speakers in 13 languages, making it versatile for various contexts such as fast-paced meetings and medical discussions.
The model excels in accuracy, ranking first on the Hugging Face Open ASR Leaderboard with a 5.33% average word error rate (WER) for English. Granary v2 enhances this model by providing a reusable data pipeline for fine-tuning, allowing developers to adapt the model to specific vocabularies and accents.
In a medical case study, fine-tuning reduced the word error rate by 45.8%, demonstrating significant improvements in transcription accuracy. This model can generate different outputs from the same audio recording based on user prompts, showcasing its flexibility and adaptability.