● Covered by 2 sources · 2 reportsMedium impact2 neutral
Nvidia Launches Nemotron 3.5 Transcribe, a Promptable Multilingual ASR Model
Key points
- Nemotron 3.5 Transcribe is an open-weight, promptable multilingual ASR model.
- It supports transcription in 40 languages and translation for 31 languages.
- The model performs speaker attribution for 13 languages and handles up to eight speakers.
- Users can provide natural language prompts to guide transcription and formatting.
- It ranks first on Hugging Face's Open ASR leaderboard for English with a 5.33% word error rate.
Nvidia Introduces Nemotron 3.5 Transcribe
Nvidia has launched Nemotron 3.5 Transcribe, its latest multilingual transcription model. This new open-weight model integrates transcription, speaker attribution (diarization), translation, and audio question answering into a single checkpoint.
The model is available on Hugging Face and ModelScope, offering developers a tool for converting recorded speech into text. It supports transcription for 40 languages, translation to or from English for 31 languages, and speaker attribution for 13 languages. It can process recordings up to an hour in length and identify up to eight speakers.
Promptable Transcription Capabilities
A key feature of Nemotron 3.5 Transcribe is its promptability, allowing users to provide specific instructions and context before the audio is processed. This means users can set the context for transcription, such as specifying that an audio file is a medical discussion in English and providing a list of medical terms.
Users can also instruct the model to label each speaker, request a verbatim or cleaned-up transcript, and define formatting for numbers. This capability aims to reduce error rates and simplify the development of applications built around these functions.
Performance and Adaptability
Nvidia states that Nemotron 3.5 Transcribe ranks first on Hugging Face’s Open ASR open model leaderboard for English, achieving an average word error rate of 5.33%. It also leads two speaker-attribution benchmarks.
The open-weight nature of the model, combined with the open Granary v2 data pipeline and fine-tuning recipes, allows developers to adapt it for specific domains and accents. This adaptability is crucial for handling diverse audio content, such as multilingual calls, fast-paced meetings, or specialized medical dictation.
Simplified Application Development
By bundling transcription, speaker attribution, translation, and audio Q&A into one checkpoint, Nemotron 3.5 Transcribe aims to simplify the work involved in building applications that utilize these capabilities. Developers can guide the output using natural language and choose where their audio is processed.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
The daily brief
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Today's brief
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's brief
New every morning, and the back catalogue is archived by date.
Primary sources
GitHub NVIDIA-NeMo/Speech GitHub NVIDIA/NeMo-Speech.cpp GitHub NVIDIA-NeMo/Curator GitHub ai-dynamo/dynamo GitHub eka-care/KARMA-OpenMedEvalKit GitHub modelscope/ms-swiftHow outlets covered it
NVIDIA released Nemotron 3.5 Transcribe, an open-weight model for multilingual speech-to-text, speaker attribution, translation, and audio Q&A. This model supports 40 languages for transcription and 31 for translation, offering developers tools to adapt it for specific domains and accents.
Nvidia released Nemotron 3.5 Transcribe, an open-weight, promptable multilingual transcription model. This model allows users to provide specific instructions and context before transcription, potentially reducing error rates and simplifying application development.