NVIDIA has introduced Nemotron 3.5 Transcribe, an advanced open-weights speech language model featuring approximately 4.3 billion active parameters. This model is designed to transcribe audio in 40 different languages, enhancing accessibility and usability in multilingual contexts.
One of the standout features of Nemotron 3.5 Transcribe is its ability to add speaker labels and word timestamps, which significantly improves the usability of transcriptions in various applications. The model has achieved a mean word error rate (WER) of 4.71% on the English Open ASR Leaderboard, demonstrating its high accuracy and reliability.
This release marks a significant step forward in the field of AI-driven transcription services, offering researchers and developers a robust tool for enhancing communication across language barriers.
Compiled automatically by the Tech AI Newsdesk from public AI-news sources and summarised in our own words.
