Sony Pixel Power calrec Sony

Now We're Talking: NVIDIA Releases Open Dataset, Models for Multilingual Speech AI

15/08/2025

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.
LINK: https://blogs.nvidia.com/blog/speech-ai-dataset-models/...
See more stories from nvidia

Most recent headlines

09/11/2025

Dalet Unveils Agentic AI Media Workflows at IBC2025

Dalet today announced a transformative leap forward for media operations: Agentic Artificial Intelligence (AI) that unifies the Dalet ecosystem under one natura...

07/11/2025

Telefnica Selects Ateme for New Cloud-Native VOD Platform

PARIS Ateme said it has been working with Spain-based multinational telco Telef nica to deploy a cloud-native media processing platform for next-generation vide...

07/11/2025

SMPTE to Hold Its Annual Meeting on Dec. 10 Via Zoom

The Society of Motion Picture and Television Engineers has announced it will hold its annual meeting on Wednesday, Dec. 10, via a Zoom webinar at 11 a.m. ET....

07/11/2025

CIMM, TVB Study Examines Local TV Measurement Problems, Solutions

NEW YORK The Coalition for Innovative Media Measurement (CIMM) and TVB have released a study examining how local viewing is measured across television, mobile d...

07/11/2025

Altice USA to Become Optimum Communications

LONG ISLAND CITY, N.Y. Altice USA announced it will change its corporate name to Optimum Communications Inc., effective Nov. 7. The company will cease trading u...

07/11/2025

Screen Australia announces Gillian Moody-Ardler as Head of First Nations Content

07 11 2025 - Media release Screen Australia announces Gillian Moody-Ardler as Head of First Nations Content Gillian Moody-Ardler. Photo credit: Sally Flegg. ...

06/11/2025

How media and information literacy in newsrooms is winning back audience trust

Media organisations from across the globe are uniting to promote media and information literacy (MIL) to help ensure their future survival and win back the trus...

06/11/2025

SVG Sit-Down: Athlos NYC CEO Kayla Green on Building the Brand of the Young Track & Field Event

SVG Sit-Down: Athlos NYC CEO Kayla Green on Building the Brand of the Young Trac...

06/11/2025

ESPN Exits ESPN BET Deal With PENN, Goes Into Business With DraftKings

ESPN Exits ESPN BET Deal With PENN, Goes Into Business With DraftKingsDraftKings will integrate innovative fan experiences' across ESPN platforms and DTC s...

06/11/2025

Peter Hujar's Day Makes the Ordinary Monumental

Ben Whishaw at the premiere of Peter Hujar's Day (photo by Robin Marshall / Shutterstock for Sundance Film Festival) By Bailey Pennick The premise is si...

06/11/2025

Spotify's Latest Feature Charts Your Week in Listening

At Spotify, we're always finding new ways to bring you closer to the music that moves you, turning sound into something personal. From the playlists that ma...

06/11/2025

Spotify Sessions chega Amrica Latina, entregando o controle criativo a trios liderados por artistas

O mundo est ouvindo a Am rica Latina. Os artistas da regi o est o constantement...

06/11/2025

Spotify Sessions Launch in Latin America, Handing Creative Control to Artist-Led Trios

The world is listening to Latin America. The region's artists are consistent...

06/11/2025

Spotify Sessions se lanza en Latinoamrica, empoderando a tros creativos liderados por artistas

El mundo est escuchando a Latinoam rica. Los artistas de la regi n marcan const...

06/11/2025

SGL Carbon: Transformation progressing rapidly

After nine months of fiscal year 2025, SGL Carbon generated sales of 652.9 million, down 16.5% on the previous year (9M 2024: 781.9 million). This is attribut...

06/11/2025

Picturepark.com is moving to fotoware.com

Contact information: Miriam Johnson, Website Lead, Fotoware Group miriam.johnson@fotoware.com Oslo, October 2025: As part of our brand unification, the Pic...

06/11/2025

Tonmeistertagung Preview

Calrec's 36-fader Argo M in Steiger Media's newly equipped OB truck parked outside the venue will showcase a live remote workflow connected to the Calre...

06/11/2025

New Gracenote Report Highlights Impact of Ineffective Content Discovery on Consumer Happiness with Streaming

With nearly 50% of streamers willing to cancel a service based on difficulty fin...

06/11/2025

ESPN Strikes Exclusive Betting Partnership With DraftKings

BRISTOL, Conn., and BOSTON ESPN and DraftKings today announced an agreement naming DraftKings the sports programmer's exclusive official sportsbook and odds...

06/11/2025

Canon Introduces New PTZ Cameras for Live Event, Studio Production

MELVILLE, N.Y. Canon USA has launched the CR-N400 and CR-N350, two new Pan-Tilt-Zoom (PTZ) cameras designed to deliver high image quality, versatile connectivit...

06/11/2025

World Series Game 7 Delivers 27.3 Million Viewers

NEW YORK Fox Sports said its coverage of the Los Angeles Dodgers' win over the Toronto Blue Jays in the decisive Game 7 of the 2025 World Series delivered 2...

06/11/2025

Roku, DoubleVerify Report Substantial Drop in Falsified Ad Impressions

NEW YORK DoubleVerify and Roku, Inc. are reporting that they have seen a marked reduction in fraudulent ad requests imitating Roku device traffic across the s...

06/11/2025

Survey: Cash-Strapped Consumers Are Cutting Costs to Keep Streaming

CAMBRIDGE, England Faced with rising inflation and worries that the economy is weakening, consumers are prioritizing their spending on popular streaming service...

06/11/2025

DPA Microphones Helps NBC Sports Let Viewers Feel' NASCAR Racing

STAMFORD, Conn. NBC Sports is mounting mics in unconventional places to bring the sounds of NASCAR into the living rooms of sports fans....

06/11/2025

Link, Aberdeen Roll Out Dual Streaming Translation, Captioning

BRANSON, Mo. Link Electronics has partnered with Aberdeen Broadcast Services to provide a real-time, dual-stream translation and captioning service....

06/11/2025

Space42 and Autonomous A2Z Establish Joint Venture to Accelerate Autonomous Mobility Transformation in the UAE

Venture will deploy and commercialize Level-4 autonomous driving technologies P...

06/11/2025

Space42 Reports 9 Months Earnings and Lays Foundation for Sustainable Growth

Strategic execution and operational optimization lay foundation for sustainable growth Space Services returns to growth following Thuraya-4 start of operations...

06/11/2025

VEON Appoints Sebastian Rice as General Counsel

06 Nov 2025 VEON Appoints Sebastian Rice as General Counsel Vitaly Shmakov promoted to the newly created Chief Investment Officer role Dubai, November 6, 2025...

06/11/2025

Telos Alliance Partners to Showcase Professional Audio Processing and Next-Gen Solutions at Tonmeistertagung

Telos Alliance Partners to Showcase Professional Audio Processing and Next-Gen ...

06/11/2025

Sky unveils first look teaser for highly anticipated prequel Gomorrah The Origins

The prequel to the Sky Original mob crime saga is a six-part drama tracking the ...

06/11/2025

Kelly Reilly and Rafe Spall star in gripping new Sky Original drama Under Salt Marsh first look teaser revealed

The six-part drama, set in a close-knit Welsh town fractured by an unspeakable c...

06/11/2025

Rohde & Schwarz at productronica 2025: Innovation and production - hand in hand

Rohde & Schwarz at productronica 2025: Innovation and production - hand in hand As a trusted partner for electronics manufacturing, Rohde & Schwarz will prese...

06/11/2025

Mystery Thriller The Price of Confession' Starring Jeon Do-yeon and Kim Go-eun Premieres December 5

Back to All News Mystery Thriller The Price of Confession' Starring Jeon D...

06/11/2025

Grass Valley Demonstrates Leadership in Hybrid and Software-Defined Production at Inter BEE 2025

The company will highlight its recent ecosystem expansions and strategic innovat...

06/11/2025

Red Seat Ventures Announces Partnership with Renowned Entrepreneur, Author and Media Personality Jillian Michaels

Red Seat Ventures Announces Partnership with Renowned Entrepreneur, Author and M...

06/11/2025

dB Broadcast Welcomes Two New Junior Systems Engineers

We're delighted to announce the appointment of Max Mehta and Mark Fannon as Junior Systems Engineers, further strengthening our growing Technology team. Ma...

06/11/2025

RT Announces Alex Walden as General Manager of the RTE Concert Orchestra

RT Announces Alex Walden as General Manager of the RTE Concert Orchestra Alex Walden, renowned orchestra manager, most recently for the BBC Concert Orche...

06/11/2025

Steve Coogan, anna Hardwicke, Holly Cairns and Kieran Cuddihy among the guests on this week's Late Late Show

Actors Steve Coogan and anna Hardwicke will be on the Late Late Show this week ...

06/11/2025

Danny O'Carroll, Anna Clifford and PJ Gallagher join the lads for THE 2 JOHNNIES LATE NIGHT LOCK IN

In the third episode of The 2 Johnnies Late Night Lock In actor Danny O'Carr...

06/11/2025

NVIDIA Founder and CEO Jensen Huang and Chief Scientist Bill Dally Awarded Prestigious Queen Elizabeth Prize for Engineering

NVIDIA founder and CEO Jensen Huang and chief scientist Bill Dally were honored ...

06/11/2025

Fall Into Gaming With 20+ Titles Joining GeForce NOW in November

Editor's note: This blog has been updated to reflect the correct launch date for Call of Duty: Black Ops 7', November 14. A crisp chill's in the...

06/11/2025

November 05, 2025

Scripps Research team identifies sugar molecules that trigger placental formation Study reveals how sugar-protein interactions are critical for the placenta dur...

05/11/2025

College Hoops Preview 2025: ESPN Remote-Ops Team Preps for Massive Slate of Men's/Women's Games

College Hoops Preview 2025: ESPN Remote-Ops Team Preps for Massive Slate of Men&...

05/11/2025

University of Iowa Centralizes Video Production With Dual Control Rooms at Carver-Hawkeye Arena

University of Iowa Centralizes Video Production With Dual Control Rooms at Carve...

05/11/2025

Ed Sheeran Brings Billions Club Live Home to Dublin

On Monday night, Ed Sheeran and Spotify lit up The Royal Dublin Society in Dublin for a one-night-only performance. The occasion? The third installment of Billi...

05/11/2025

Cumbia Takes Center Stage as Spotify Celebrates Argentina's Most-Listened-To Genre

Cumbia has long been woven into daily life in Argentina, and its popularity on S...

05/11/2025

La cumbia toma el centro del escenario mientras Spotify celebra el gnero ms escuchado de Argentina

La cumbia forma parte del d a a d a de los argentinos desde siempre, y su popula...

05/11/2025

Spotify Publishes First Independent Author Releases Through Audiobook Selects, With More Planned Ahead

Earlier this year, our in-house publishing imprint, Spotify Audiobooks, put out ...

05/11/2025

SBS showcases next generation of pro cyclists in extended broadcast deal with ProVelo Super League

SBS showcases next generation of pro cyclists in extended broadcast deal with Pr...