Nvidia Unveils Nemotron 3.5 Transcribe: An Open Transcription Model for Multilingual Audio

Image: TechCrunch · Source

Nvidia launches Nemotron 3.5 Transcribe, a free open-weight speech-to-text model supporting up to 40 languages and distinguishing multiple speakers, optimized for recorded audio transcription in business and medical settings.

On October 28, 2026, Nvidia announced Nemotron 3.5 Transcribe, a new free and open-weight speech-to-text model capable of transcribing recordings in up to 40 languages. Alongside the model, Nvidia released version 2 of Granary, an updated open-source speech dataset aimed at improving transcription quality across diverse speech samples.

Nemotron 3.5 can differentiate up to eight speakers within a recording and translate 31 languages to and from English. It supports handling recordings up to an hour long, catering primarily to developers focused on building transcription applications for business meetings and medical contexts—two sectors where AI-driven speech-to-text technologies have become increasingly valuable.

A key advantage of Nemotron 3.5 is its ability to outperform competitors on multiple benchmark tests, including Hugging Face's Open ASR Leaderboard. Nvidia emphasizes the model's flexibility for customization through its NeMo Speech toolkit, enabling apps to recognize specialized jargon specific to different industries and languages.

Nemotron 3.5 is notable for its efficiency, encompassing 30 billion parameters yet running on a single chip. It is compatible with a range of Nvidia hardware, including the Jetson AGX Thor robotics controller, the DGX Spark chip featured in new laptops, and the Blackwell data center processors.

However, Nemotron 3.5 is designed exclusively for recorded audio and does not support live audio streams, leaving real-time transcription use cases to other providers such as ElevenLabs. Nvidia's initiative joins a broader surge of AI voice technology releases recently, contemporaneous with products like Mistral’s Voxtral 3, ElevenLabs’ Realtime voice agents, and Salesforce’s Launa-Omni voice model.

Overall, Nvidia's release represents a significant advancement in accessible, multilingual speech transcription technology tailored for offline and recorded audio applications in professional environments, supporting further innovation in AI-powered voice tools.

Category: Apps Release_status: launched Release_date: 2026-10-28 Product_name: Nemotron 3.5 Transcribe