Observed Signal · Oct 28, 2026 · Technical Release · Source: techcrunch · Impact: 4/5 · Sentiment: Positive
Nvidia Releases New Open Transcription Model
Nvidia has released Nemotron 3.5 Transcribe, a free, open-weight speech-to-text model that can transcribe audio in up to 40 languages and translate 31 languages to and from English. It supports recordings up to one hour long and can distinguish up to eight speakers. The company also released version 2 of its Granary open-source speech dataset. The model is targeted at developers building meeting transcription and medical transcription applications. It runs on a single chip, from edge devices like Jetson AGX Thor to datacenter Blackwell chips. Nvidia claims it beats rivals on benchmarks like Hugging Face's Open ASR Leaderboard. However, it only works with recorded audio, not live streams, leaving real-time use to competitors like ElevenLabs. The release comes amid a wave of voice AI model launches from Mistral, ElevenLabs, and Salesforce.
This is a significant release from a major tech company like Nvidia, providing an open-weight transcription model that can drive innovation in meeting and medical transcription tools. It strengthens Nvidia's position in the AI infrastructure ecosystem, potentially impacting how businesses build and deploy speech AI applications.
Track NVIDIA Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your Polaris7 workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Nvidia released Nemotron 3.5 Transcribe, an open-weight speech-to-text model supporting 40 languages.
- The model can translate 31 languages to and from English and handle recordings up to one hour long.
- Nvidia also released version 2 of its Granary open-source speech dataset.
- The model runs on a single chip, including Jetson AGX Thor and DGX Spark chips.
- The model only works with recorded audio, not live streams.
Connected Companies & Entities
4 Entities mapped“Nvidia on Wednesday released Nemotron 3.5 Transcribe, a free, open-weight model that transcribes audio in up to 40 languages.”
“ElevenLabs launched Realtime for instant speech in voice agents.”
“Salesforce released Launa-Omni, its voice model.”
“Mistral shipped Voxtral 3 earlier this month.”
Ontology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Cohere open-sources Transcribe voice model
Cohere has released Transcribe, its first open-source automatic speech recognition (ASR) model designed for transcription tasks like note-taking and speech analysis. The 2-billion-parameter model is optimized for consumer-grade GPUs and supports 14 languages. Cohere says Transcribe achieved a 5.42 average word error rate (WER) on the Hugging Face Open ASR leaderboard, outperforming several comparator models, and processed audio at an estimated 525 minutes per minute. Human evaluators favored its outputs in head-to-head comparisons (61% average win rate), though it lagged on Portuguese, German and Spanish. Cohere will offer Transcribe free via its API, make it available on Model Vault, and plans integration into its enterprise orchestration platform North.
Building Inflection AI’s Multi-Speaker Capabilities with NVIDIA Nemotron Speech Open Models
Inflection AI announced a new blog post detailing its multi-speaker capabilities built with NVIDIA Nemotron Speech Open Models, published on Oct 28, 2026.
Microsoft Launches MAI-Transcribe-2, Cheapest and Most Accurate in Market
Microsoft AI has launched MAI-Transcribe-2, a speech-to-text model that supports 60 languages and achieves top accuracy on the FLEURS benchmark with a 5.2% average word error rate. The model is 10x faster than OpenAI's GPT-Transcribe, 7x faster than ElevenLabs' Scribe v2, and 5x faster than Google's Gemini 3.5 Transcribe. It offers features like speaker diarization, word-level timestamps, keyword biasing, and code-switching. Priced at $0.10 per audio hour (promotional until end of year), it undercuts competitors significantly. For Thai, it achieves a 3.4% word error rate, besting Gemini 3.5 Transcribe (3.8%) and Whisper v3-large (8.7%). The model is available via Microsoft Foundry, MAI Playground, and OpenRouter, and is part of Microsoft's strategy to replace OpenAI technologies with in-house models.
Receive all future signals for NVIDIA & competitors in your briefing
Never miss material M&A deals, leadership changes, account wins, or product rollouts. Polaris7 monitors 20,000+ primary sources 24/7, delivering verified telemetry directly into your executive briefing.