Meta has developed a new model of artificial intelligence, with which users will be able to translate speech and text materials into almost 100 languages.
The new configuration of digital intelligence, created by a team of specialists of the technology giant, was named SeamlessM4T. This development is a universal multimodal and multilingual translation model using AI. The relevant information is contained in the company’s press release, which was published last Tuesday, August 22.
The new model of machine intelligence supports speech recognition, translation of speech into text, translation of information existing in a verbal format into an oral statement in another language, translation of text material into a similar form of the content of thoughts and conclusions in another language, and translation of words spoken into words written.
The press release of the technology giant indicates that the advantage of the new AI configuration is the algorithm of its functioning, providing for the application of a systematic approach to solving tasks. The company claims that this principle of operation increases the level of efficiency of the machine intelligence model and contributes to ensuring an acceptable quality indicator. The technology giant stated that its new product allows for a minimum number of errors and carries out the process of translating information materials existing in oral and written formats in the shortest possible time.
SeamlessM4T was publicly released under a research license. Information about this is contained in the company’s press release. The license allows researchers and developers to use all the features of the AI model.
The company also publishes the metadata of SeamlessAlign, an open data set of multimodal translation, which includes 270 thousand hours of processed speech and text alignment. This set of information materials will become a resource for new developments and research in the area of AI.
SeamlessM4T is based on the previous achievements of the technology giant concerning the development of language translation technologies. Last year, the company released a text-to-text machine translation model called No Language Left Behind (NLLB), which supports 200 languages. This AI configuration has been integrated into Wikipedia as a translation provider.
Meta created Universal Speech Translator, the first direct speech-to-speech translation system for Hokkien, a type of Chinese language that does not have a developed writing system. The company has also released a multilingual Speech program that offers users speech recognition and synthesis technologies and also has a language identification function.
As we have reported earlier, Meta Introduces AI Model With Speech Generation Capabilities.