Jonathan Cohen

Career

NVIDIA's author pages describe Cohen as having spent eight years at NVIDIA in engineering and research roles, then three years at Apple as a Director of Engineering, before returning to NVIDIA. There Cohen became a Senior Director for AI Software, leading development of NVIDIA's speech and language tools and technologies, including the NeMo framework.¹²

NVIDIA's December 2025 Nemotron 3 white paper lists Cohen among the leadership of the Nemotron 3 model family, and the April 2026 Nemotron 3 Super technical report lists Cohen among its contributors.³⁴

Products and projects

Cohen's posts on NVIDIA's blogs trace the speech and language software Cohen's teams have released. A 2019 technical post explained how to build domain-specific automatic speech recognition models on GPUs, and a 2023 post announced NeMo Guardrails, open-source software for keeping generative AI chatbot responses on track.²¹

In August 2025 Cohen announced Granary, an open multilingual speech dataset, released together with the Canary-1b-v2 and Parakeet-tdt-0.6b-v3 speech models to extend speech AI to more languages.⁵

Work and contributions

• NVIDIA NeMo Open-source code

  NVIDIA's blog profile says Cohen led development of NVIDIA's tools and technologies for speech and language, including NeMo; Cohen also wrote NVIDIA's April 2023 announcement of the open-source NeMo Guardrails software.¹²

• NVIDIA Nemotron Open weights

  Listed in the leadership credits of the NVIDIA Nemotron 3 white paper (December 2025), which states that NVIDIA will release the model weights, training datasets and recipes; also listed among the contributors to the Nemotron 3 Super technical report (April 2026).³⁴

• Granary dataset with Canary and Parakeet models Open data

  Wrote NVIDIA's August 2025 announcement of Granary, an open-source corpus of multilingual speech audio, released on Hugging Face with the Canary-1b-v2 and Parakeet-tdt-0.6b-v3 models trained on it.⁵

Timeline

1. Dec 17, 2019Wrote an NVIDIA technical post on building domain-specific speech recognition models on GPUs.²
2. Apr 25, 2023Announced NeMo Guardrails, open-source software for adding guardrails to AI chatbots.¹
3. Aug 15, 2025Announced the Granary multilingual speech dataset and the Canary and Parakeet models built with it.⁵
4. Dec 2025Listed in the leadership credits of NVIDIA's Nemotron 3 white paper.³
5. Apr 2026Listed among the contributors to the Nemotron 3 Super technical report.⁴

Other documented roles

Roles the sources document that are not confirmed as current.

• Senior Director for AI Software, in charge of speech and language tools including NeMo, NVIDIA¹²
• Director of Engineering, Apple¹²

Where they work

• NVIDIA 43 people in the wiki

Related people

• Rama AkkirajuAlso at NVIDIA
• Clément FarabetAlso at NVIDIA
• Frédéric BastienAlso at NVIDIA
• Alexander AminiAlso at NVIDIA
• Bratin SahaAlso at NVIDIA
• Gabriel SynnaeveLanguage models · Speech and audio
• Ronan CollobertML frameworks and libraries · Speech and audio
• Dong YuSpeech and audio · ML frameworks and libraries
• Michael AuliSpeech and audio · Language models
• Koray KavukcuogluLanguage models · Speech and audio