Nvidia is in a delicate situation caused by the conflict between the United States and China, which have led her to the edge of the abyss on more than one occasion and that seems to have no end, despite the fact that the Trump administration has already given her green light and green light and permission to put the H20 chips into circulation.
Xi Jinping’s response, on the other hand, was not expected, since it quickly decided malwarewhat Nvidia quickly denied.
While shortly after he advanced that he was working on «a variety of products» to deal with these problems in China, sources in this country had already indicated that they would prioritize national products against foreigners, a measure that will probably promote Huawei competitor.
This instability has led Nvidia to think of a plan B and, instead of looking towards Asia, has chosen to focus the focus on the West and, more specifically, in the countries of Europe. Thus, he has launched a new set of data and models that support The development of the AI of voice recognition and translation High quality for 25 European languages.
Technology for multilingual chatbots and simultaneous translation services
The technological directed by Jensen Huang wanted to address an important problem related to the ability of artificial intelligence, since «of the approximately 7,000 languages that exist in the world, only a small fraction has the support of linguistic models» driven by this technology.
By making available to users these free use tools, therefore, he hopes that developers can «climb the AI applications more easily to support global users» with technologies aimed at platforms such as chatbots, customer service agents and translation services in real time.
Granary, meanwhile, contains about one million hours of audio, including almost 650,000 hours for voice recognition and more than 350,000 for translation. With these data, developers will be able to build models that address transcription and translation tasks in almost All 24 official languages of the European Union, in addition to the Ukrainian and Russian.
Also, Nvidia Canary -1b -V2 is a model of one billion parameters trained in Granary for the high quality transcription of European languages, in addition to the translation between English and two dozen compatible languages.
This model is available under a permissive license and expands the languages admitted by the Canary family from four to 25. Also, according to your data, it is able to offer a transcription quality and translation comparable to three larger models and executes the inference at a speed 10 times higher.
NVIDIA PARKEET-TDT-0.6B-V3, on the other hand, contains 600 million parameters and is designed for real-time transcription or large volumes of languages compatible with Granary, as explained in an official statement.
It has also nuanced that this model prioritizes high performance and is capable of transcribing 24 -minute audio segments in a single pass of inference. It also detects the language of the input audio automatically and transcribes it without additional instructions.
With all these novelties, for which Nvidia has collaborated with researchers from the Carnegie Mellon University of Pennsylvania (United States) and Fondazione Bruno Kessler of Trent (Italy), the technological brand wanted to demonstrate that it wants to cover more inclusive and universal voice processing technologies.
Know How we work in NoticiasVE.