Of around 7,000 languages in the world, a tiny fraction are supported by AI language models. NVIDIA is tackling the problem with a new dataset and models that support the development of high-quality speech recognition and translation AI for 25 European languages - including languages with limited available data like Croatian, Estonian and Maltese.
These tools will enable developers to more easily scale AI applications to support global users with fast, accurate speech technology for production-scale use cases such as multilingual chatbots, customer service voice agents and near-real-time translation services. They include:
Granary, a massive, open-source corpus of multilingual speech datasets that contains around a million hours of audio, including nearly 650,000 hours for speech recognition and over 350,000 hours for speech translation.
NVIDIA Canary-1b-v2, a billion-parameter model trained on Granary for high-quality transcription of European languages, plus translation between English and two dozen supported languages. It tops Hugging Face's leaderboard of open models for multilingual speech recognition accuracy.
NVIDIA Parakeet-tdt-0.6b-v3, a streamlined, 600-million-parameter model designed for real-time or large-volume transcription of Granary's supported languages. It has the highest throughput of multilingual models on the Hugging Face leaderboard, measured as duration of audio transcribed divided by computation time.
The paper behind Granary will be presented at Interspeech, a language processing conference taking place in the Netherlands, Aug. 17-21. The dataset, as well as the new Canary and Parakeet models, are now available on Hugging Face.
How Granary Addresses Data Scarcity To develop the Granary dataset, the NVIDIA speech AI team collaborated with researchers from Carnegie Mellon University and Fondazione Bruno Kessler. The team passed unlabeled audio through an innovative processing pipeline powered by NVIDIA NeMo Speech Data Processor toolkit that turned it into structured, high-quality data.
This pipeline allowed the researchers to enhance public speech data into a usable format for AI training, without the need for resource-intensive human annotation. It's available in open source on GitHub.
With Granary's clean, ready-to-use data, developers can get a head start building models that tackle transcription and translation tasks in nearly all of the European Union's 24 official languages, plus Russian and Ukrainian.
For European languages underrepresented in human-annotated datasets, Granary provides a critical resource to develop more inclusive speech technologies that better reflect the linguistic diversity of the continent - all while using less training data.
The team demonstrated in their Interspeech paper that, compared to other popular datasets, it takes around half as much Granary training data to achieve a target accuracy level for automatic speech recognition (ASR) and automatic speech translation (AST).
Tapping NVIDIA NeMo to Turbocharge Transcription The new Canary and Parakeet models offer examples of the kinds of models developers can build with Granary, customized to their target applications. Canary-1b-v2 is optimized for accuracy on complex tasks, while parakeet-tdt-0.6b-v3 is designed for high-speed, low-latency tasks.
By sharing the methodology behind the Granary dataset and these two models, NVIDIA is enabling the global speech AI developer community to adapt this data processing workflow to other ASR or AST models or additional languages, accelerating speech AI innovation.
Canary-1b-v2, available under a permissive license, expands the Canary family's supported languages from four to 25. It offers transcription and translation quality comparable to models 3x larger while running inference up to 10x faster.
https://blogs.nvidia.com/wp-content/uploads/2025/08/Canary-demo.mp4
NVIDIA NeMo, a modular software suite for managing the AI agent lifecycle, accelerated speech AI model development. NeMo Curator, part of the software suite, enabled the team to filter out synthetic examples from the source data so that only high-quality samples were used for model training. The team also harnessed the NeMo Speech Data Processor toolkit for tasks like aligning transcripts with audio files and converting data into the required formats.
Parakeet-tdt-0.6b-v3 prioritizes high throughput and is capable of transcribing 24-minute audio segments in a single inference pass. The model automatically detects the input audio language and transcribes without additional prompting steps.
Both Canary and Parakeet models provide accurate punctuation, capitalization and word-level timestamps in their outputs.
Read more on GitHub and get started with Granary on Hugging Face.
Most recent headlines
11/12/2025
Dalet, a leading provider of cloud-native, end-to-end media workflow solutions, ...
06/12/2025
In a live broadcast from the Reagan National Defense Forum, L3Harris Chair and CEO Christopher Kubasik joined Morgan Brennan on CNBCs Closing Bell: Overtime. Ku...
06/12/2025
FORT LAUDERDALE, Fla. A new survey from Pixitmedia by Datacore revealed a major shift in the Media & Entertainment industry in media archiving, with 85% of resp...
06/12/2025
HACKENSACK, N.J. LiveU has announced that the national public broadcaster Czech Television has completed one of the largest LiveU live production deployments fo...
06/12/2025
NEW YORK The National Academy of Television Arts and Sciences (NATAS) presented the Excellence in Production Technology Emmy Award to NASA+ and Dr. Tom Leight...
05/12/2025
2025 Sports Broadcasting Hall of Fame: Curt Gowdy Jr. - Master Storyteller, Nati...
05/12/2025
SVG Sit-Down: Veritone's Sean King on the Power of Mining Video, Audio DataThe company's Data Refinery offers users total control and governance over da...
05/12/2025
Platinum White Paper: Inside the Nashville Predators' Unified, Flexible, Sca...
05/12/2025
Netflix Reaches Agreement To Acquire Warner Bros. Following Planned WBD SplitThe deal does not include WBDs sports assets like TNT Sports (US, UK, LatAm), Euros...
05/12/2025
FOX Sports Returns to Indianapolis for Primetime Broadcast of Big Ten Championsh...
05/12/2025
SVG Summit 2025 Preview: Digital Engagement & Monetization Workshop Tackles the ...
05/12/2025
Atlanta United Lights Up New Emory Healthcare Studio With First Live Broadcast f...
05/12/2025
As Messi Takes the Pitch, MLS, Apple, NEP Roll Out Largest MLS Cup Production Ev...
05/12/2025
ESPN Enters College Football's Most Intense Month With Elevated Workflows fo...
05/12/2025
It's about that time! Awards season is in full swing, and the Film Independe...
05/12/2025
Every year, Spotify Wrapped offers a personalized look back at the audio that defined your year. It's a snapshot of your listening habits, designed to tell ...
05/12/2025
In 2025, Spotify's EQUAL, GLOW, and RADAR programs celebrated women, LGBTQIA , and emerging artists who turned moments into milestones. From breaking record...
05/12/2025
In our latest blog, we explain how Wi-Fi 7 rollouts can drive consumer loyalty with value-add services such as consumer cybersecurity. We also explore how this ...
05/12/2025
LOS ANGELES Netflix announced it has entered into an agreement to acquire the assets of Warner Bros. for $82.7 billion....
05/12/2025
NEW YORK Nielsens Gracenote has launched Gracenote Content Connect, a new ad platform that provides agencies, brands, supply-side platforms (SSPs) and demand-si...
05/12/2025
NEW YORK In an most important update to the workings of deal-based programmatic advertising, IAB Tech Lab has released version 1.0 of its Deals API for public c...
05/12/2025
NEW YORK Pass the turkey. Pass the stuffing. Pass the cranberry sauce. All are common requests of Americans celebrating Thanksgiving Day with family and f...
05/12/2025
NEW YORK Iris, the new cloud-connected camera control platform, has officially launched with features that turn virtually any PTZ camera into a software-connect...
05/12/2025
HOLLYWOOD, Calif. Netflix announced today that it has entered into an agreement to acquire the assets of Warner Bros. for $82.7 billion....
05/12/2025
NEW YORK Iris, the new cloud-connected camera control platform, has officially launched with features that turn virtually any PTZ camera into a software-connect...
05/12/2025
WASHINGTON The Federal Communications Commission has approved AT&T's $1.02 billion acquisition of spectrum from UScellular in a decision that was issued sho...
05/12/2025
The Best Coldplay Songs: 21 Tracks That Shoot for the Stars From Yellow to Viva La Vida, Fix You to Paradise, this playlist goes back to the start.
December ...
05/12/2025
Zafris Lecture Series Brings Nabil Ayers to Berklee The 32nd annual James G. Zafris Distinguished Lecture series was held on Thursday, November 13, with guest...
05/12/2025
Back to All News
Netflix Unveils the First Look for Land of Sin' - Premier...
05/12/2025
Introducing New Perks to Help You Get Even More from LinkedIn Premium Published on Dec 5, 2025 Categories: Company News, Product News
LinkedIn Corporate Co...
05/12/2025
Friday 5 December 2025
A new Game of Thrones Tale: Official trailer for Sky Exc...
05/12/2025
Back to All News
Don Lee, Lee Jin-uk, and Lalisa Manobal to Star in Netflix Act...
05/12/2025
Tis the season of giving once again and this year we've taken our Give Back Fridays' concept and turned it on its head.
In the autumn we were approach...
05/12/2025
Brayden Gogis doesn't remember a time when he wasn't completely fixated on games in all forms. In preschool, when they asked us to dress up as what we ...
05/12/2025
The Grinch steals the spotlight as the theme for The Late Late Toy Show 2025
Tune in tonight at 9:35pm on RT One and worldwide on RT Player
#LateLateToyShow...
05/12/2025
RT Announces New Presenters of Flagship News Programmes
New RT Six One News co-presenter Tommy Meskill
Sarah McInerney & Justin McCarthy join Morning Ir...
04/12/2025
ToolsOnAir Blackmagic Design HyperDeck Event Presets for just:in mac pro 2025 & ...
04/12/2025
ToolsOnAir AJA Ki Pro Event Presets for just:in mac pro 2025 & just:in linux
More Details:Starting with version 5.5, both just:in mac pro and just:in linux sol...
04/12/2025
Wangu Kanuri from Kenya and Godwin Asediba from Ghana are two of this years finalists for Thomsons Young Journalist of the Year Award. The pair are runners-up i...
04/12/2025
SVG Sit-Down: ProximaVision's Claudio Lisman on Why Tethered Drones Could Be...
04/12/2025
SVG Campus Shot Callers: Imry Halevi, Senior Associate Director of Athletics, Co...
04/12/2025
Platinum White Paper: LiveU Lightweight Sports Production: A Step Change in Spor...
04/12/2025
London to Riyadh: DAZN brings the boxing glamour to new production levels for Be...
04/12/2025
Analysis: Paramount bets on the battering ram' with Champions League play By Callum McCarthy, Editor-at-Large
Tuesday, December 2, 2025 - 10:12
Print ...
04/12/2025
Space City Home Network Launches SCHN DTC App for Astros and RocketsThe Rockets and Astros were previously the lone NBA and MLB teams without a DTC appBy Jason...
04/12/2025
SVG Summit 2025 Preview: Content Workflows Workshop Spotlights Evolution of Spor...
04/12/2025
New Sponsor Spotlight: Geotech's Patrick Wambold On the Unreal Engine Revolu...
04/12/2025
Curt Gowdy Jr. - Master Storyteller, Nationally and RegionallyBy Jason Dachman, Editorial Director, U.S.
Thursday, December 4, 2025 - 1:52 pm
Print This Sto...
04/12/2025
(L-R) Rebecca Lichtenfeld, Mohammadreza Eyni, Sara Khaki, and Judith Helfand att...
Of around 7,000 languages in the world, a tiny fraction are supported by AI language models. NVIDIA is tackling the problem with a new dataset and models that support the development of high-quality speech recognition and translation AI for 25 European languages - including languages with limited available data like Croatian, Estonian and Maltese.
