Nvidia has announced the release of Nemotron 3.5 Transcribe, a new open-weight audio transcription model capable of processing speech across up to 40 languages. Available for free, the 30-billion-parameter model represents Nvidia’s latest push into enterprise-grade speech recognition, arriving alongside version 2 of Granary, the company’s open-source speech dataset. The launch positions Nvidia directly within a rapidly accelerating voice AI market, providing developers and enterprises with powerful tools tailored for high-demand workplace applications such as automated meeting minutes and medical dictation. Read Also: The Battle Over AI Safety: Silicon Valley Giants Clash With Washington as Regulatory Pressures Mount How Instinct Became the Buzzy AI Agent Taking on Big Tech Without a Mascot According to Nvidia, Nemotron 3.5 Transcribe is designed to handle complex audio environments with remarkable precision. The model boasts the ability to translate 31 languages both to and from English, process extended recordings lasting up to a full hour, and accurately perform speaker diarization—the process of partitioning an audio stream into speaker-homogeneous segments—for up to eight distinct individuals in a single recording. Nvidia highlights that the model performs competitively against industry rivals, claiming top-tier results on established evaluation benchmarks, including Hugging Face’s Open ASR Leaderboard. The primary target audience for Nemotron 3.5 Transcribe includes software developers and enterprise engineering teams building applications for business meetings and medical transcription. These two sectors represent some of the earliest and most firmly established commercial footholds for artificial intelligence in the modern workplace. To help developers tailor the technology to specific operational requirements, Nvidia is offering a comprehensive suite of resources. Alongside the new model and the updated Granary dataset, developers can utilize tools like Nvidia’s NeMo Speech framework. This ecosystem allows organizations to fine-tune and customize applications so they can accurately recognize specialized industry jargon, technical terminology, and proprietary acronyms across multiple languages. Despite its robust feature set, the model does come with certain operational constraints. Most notably, Transcribe is currently engineered to work with recorded audio files rather than live, real-time audio streams. This design choice aligns with its primary use cases in retrospective documentation, such as reviewing recorded corporate conferences or transcribing completed patient consultations. From a hardware perspective, Nvidia has emphasized the deployment flexibility of Nemotron 3.5 Transcribe. Despite scaling at 30 billion parameters, the company promises that the model is optimized to run efficiently on a wide variety of single-chip configurations. This versatility spans Nvidia’s entire hardware ecosystem, ranging from the Jetson AGX Thor robotics controller and the DGX Spark chip designed to power a new generation of Windows 11 AI-enabled laptops, all the way up to enterprise-grade Blackwell datacenter chips. With this release, Transcribe joins Nvidia’s growing portfolio of open-weight voice models. It complements smaller yet highly efficient offerings such as Parakeet, a multilingual automatic speech recognition model operating at 1.1 billion parameters. Nvidia’s introduction of Nemotron 3.5 Transcribe coincides with a remarkably busy period for generative AI voice technology releases across the tech sector. The competitive landscape has seen a flurry of activity in recent weeks, highlighting the intense industry focus on human-computer voice interaction and automated speech processing. Earlier this month, Mistral rolled out its Voxtral 3 voice model, while ElevenLabs expanded its ecosystem with the launch of Realtime, a solution designed for instant speech generation in conversational voice agents. Additionally, Salesforce recently entered the fray with the release of Launa-Omni, its own proprietary voice model targeting enterprise workflows. As enterprises increasingly demand sophisticated, multilingual, and accurate speech capabilities to manage the massive influx of digital communication and documentation, Nvidia’s open-weight approach provides developers with the underlying infrastructure needed to deploy customized voice solutions at scale. Post navigation Jio Platforms Surges 21% in Stock Market Debut Following India’s Largest IPO
