Goal has announced the launch of SeamlessM4Ta multimodal AI model that works with both text and speech translations and that it intends to become a step towards the creation of a universal translator. It is, according to the company, a neural network that can process both text and sound, and that is capable of translating between up to a hundred languages. Also to do it text to speech, voice to text, voice to voice or text to text. Its goal is to help people who speak different languages to communicate with each other more effectively.
The company has released the model with a research license, which allows developers to use it as the basis for their work and development. Among its functions, the model can perform voice recognition tasks, and if you provide it with audio of spoken phrases and texts, it can convert it to text. In addition, it is capable of automatically recognizing the languages that are provided to it in text or voice for translation.
It is also capable of performing translations at the same time that it transcribes audio and turns it into text. And if you provide an audio clip with words and phrases, it can translate them into another language, in voice. Of course, it is capable of doing text-to-text translations, much like Google Translate does. Of course, the text translation functions support, as we have mentioned, around 100 languages. But in voice translations, its number of supported languages is reduced to 35 (English included).
In addition to the model, Meta has confirmed that will cast the project metadata into a datasetwho has called SeamlessAlign. According to those responsible, it is the largest open source multimodal data set released to date. It contains 270,000 hours of spoken speech and text settings, which are what have been used to train the model.
Its availability means that in the future, the training of Artificial Intelligence models for translation will be easier and more agile for researchers. For more information on SeamlessM4T it is worth visiting the Meta GitHub page dedicated to research in this area.
