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
09/11/2025
Dalet today announced a transformative leap forward for media operations: Agentic Artificial Intelligence (AI) that unifies the Dalet ecosystem under one natura...
15/10/2025
Across the world, early-career reporters are interrogating systems that shape daily life, from mortuaries and classrooms to the corridors of legal education and...
15/10/2025
St. Louis Blues Pay Tribute to Ozzy Osbourne in New Pregame Open VideoThe NHL franchise worked with production firms 2WEI and 4th Floor CreativeBy Mark J Burns,...
15/10/2025
Cobalt Digital, G3DVu, Proxima Vision, and Telos Alliance Renew Premier Sponsors...
15/10/2025
By Paige Bethmann
Before my grandmother passed away in 2019, I went to visit her in the hospital where she'd been for a few weeks. When I arrived, the nurs...
15/10/2025
Two years ago, Spotify set out to grow the entire publishing industry by making ...
15/10/2025
People turn to DJ for a listening experience that feels personal, dynamic, and j...
15/10/2025
Las personas recurren a DJ para vivir una experiencia de escucha personal, din m...
15/10/2025
Rock legend Bruce Springsteen and luminary actor Jeremy Allen White came togethe...
15/10/2025
At Spotify, we're all about supporting rising talent and helping them connec...
15/10/2025
New SBS Documentary Series THE CANCER KILLERS Premieres 4 November on SBS & SBS ...
15/10/2025
SBS celebrates the Festivals of Lights with bold, illuminating stories across th...
15/10/2025
One of humanity's most profound questions continues: Is there any place in our galaxy suitable for life beyond Earth? NASA intends to find out, with the hel...
15/10/2025
eds3_5_jq(document).ready(function($) { $(#eds_sliderM519).chameleonSlider_2_1({ content_source:......
15/10/2025
Scality, a global leader in cyber-resilient storage software for the AI era, today announced the advancement of its comprehensive AI ecosystem certification pro...
15/10/2025
The Hollywood Professional Association (HPA) today unveiled key highlights of the 2026 HPA Tech Retreat, scheduled for Feb. 15 19 at the Westin Rancho Mirage Go...
15/10/2025
At NAB Show New York 2025 (Stand 244), Interra Systems will showcase the future of media QC, monitoring, and captioning, highlighting its award-winning, AI- and...
15/10/2025
Grass Valley today announced that RMC BFM, part of the CMA-CGM Group and France's third-largest private media group, has chosen Grass Valley's productio...
15/10/2025
Appear, the global leader in live production technology, will demonstrate how its powerful hardware and software innovations are shaping the future of hybrid, s...
15/10/2025
Rakim to Join Berklee's Hip-Hop Hall of Fame at Signature Series Concert The rap icon becomes the latest inductee, following luminaries such as Roxanne Sh...
15/10/2025
Berklee Brings Live Music to the Head of the Charles Student artists will perform on multiple riverfront stages during the 60th anniversary of the world's...
15/10/2025
CobbTV, the government access television channel for Cobb County, Ga. recently acquired an automated playout system from Pebble, a global provider of automation...
15/10/2025
OSLO, Norway Appear, the global leader in live production technology, will demonstrate how its powerful hardware and software innovations are shaping the future...
15/10/2025
NEW YORK The NBA is making major changes to the NBA App and NBA TV as it takes control of them from TNT Sports, which has long managed the league's digital ...
15/10/2025
SAN MATEO, Calif. In what promises to be a major expansion of interactive features and personalized content on the DirecTV platform, the operator and Glance hav...
15/10/2025
SAN JOSE, Calif. Roku has launched changes to its user interface (UI) that the streaming platform says will better showcase original programming on the platform...
15/10/2025
LOS ANGELES Software-defined data storage and data services provider OpenDrives has elevated Alex Dunfey to chief technology officer, responsible for driving th...
15/10/2025
Abu Dhabi, UAE October 15, 2025: Space42 (ADX: SPACE42), the UAE-based AI-powe...
15/10/2025
Sheldon Nichols and Will Trickett to find and prepare more classic cars for budget conscious would-be owners from Wiser Films
UKTV have recommissioned the clas...
15/10/2025
Wednesday 15 October 2025
Nick Cave
JPEG (275 KB)
Explore the creative univer...
15/10/2025
Wednesday 15 October 2025
From breaking up with your partner to quitting your job - new polling shows some messages are just too scary to send
Over a fifth (...
15/10/2025
Rohde & Schwarz unveils compact MXO 3 oscilloscopes with 4 and 8 channels: Advan...
15/10/2025
Back to All News
Revenge Series The Resurrected' Captivates Audiences Across Asia
Entertainment
15 October 2025
GlobalTaiwan
Link copied to clipboard
...
15/10/2025
Back to All News
Bringing the Best in VFX and Virtual Production Together as Eyeline
Jeffrey Shapiro
CEO, Eyeline
Business
15 October 2025
GlobalCanadaInd...
15/10/2025
Deployed by Astound Business Solutions, Harmonic's Primary Distribution Solution Ensures Outstanding Video Quality and Seamless Ad Insertion at the Edge
SA...
15/10/2025
Collaboration extends a trusted relationship as RMC BFM invests in scalable, fut...
15/10/2025
Series coming in 2026 stars Tom Vaughan-Lawlor, Justine Mitchell and Jason O'Mara released today
RT today released first look images of new comedy-drama ...
14/10/2025
SVG Europe Summit 2025: All Sessions Now Available to Watch on SVG PLAYNetworking event that preceded IBC2025 shone a light on elite live sports innovation acro...
14/10/2025
SVG Sit-Down: Author Rich Podolsky on Writing Madden & Summerall: How They Revo...
14/10/2025
SVG All-Stars: Michael Reiners, Coordinating Producer, FloRacingThe Illinois State grad steers a vast schedule of motorsports events at tracks across the countr...
14/10/2025
Content protection: Getting the right management for your DRM By Neal Romanek
Friday, October 10, 2025 - 10:11
Print This Story
Eluvio power the EPCR'...
14/10/2025
As League Takes Over Ops, NBA TV and NBA App Add 60 Games, Weekday Studio Show, ...
14/10/2025
Time and effort: World's largest student-led broadcast prepares to go On Air...
14/10/2025
(L-R) Guest, Kimberly Robinson Jones, Geeta Gandbhir, Pamela Dias, and Takema Ro...
14/10/2025
Lossless ist jetzt mit Spotify Premium verf gbar.
Verlustfreies Audio war eine...
14/10/2025
La qualit Lossless est disponible sur Spotify Premium.
Le format sans perte de...
14/10/2025
For the seventh edition of Spotify and FC Barcelona's artist jersey series, ...
14/10/2025
Spotify is committed to bringing the best listening experience to all our users, and that includes parents and families. That's why we're expanding mana...
14/10/2025
Since its debut, the Spotify Original podcast Caso 63 has been more than just a story; it's been a cultural sensation. The science fiction thriller captivat...
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.
