Meta Platforms introduces the SeamlessM4T AI model, facilitating multilingual speech translation and transcription, aligning with its vision of global interactions within the metaverse. The model supports text-to-speech translations for almost 100 languages and full speech-to-speech translation for 35 languages, combining capabilities from separate models, while being offered for non-commercial use.
23 August 2023 – In a significant stride toward bridging global communication barriers, Meta Platforms, the parent company of Facebook, has unveiled its latest AI advancement – the SeamlessM4T model. This cutting-edge AI model possesses the remarkable ability to accurately translate and transcribe speech across a multitude of languages, presenting a potential solution for real-time cross-lingual interactions.
Meta Platforms announced through an official blog post that the SeamlessM4T AI model showcases the capability to seamlessly facilitate translations between text and speech in nearly 100 languages. Additionally, it empowers users with full speech-to-speech translation for 35 languages, marking a substantial advancement that amalgamates functionalities which were hitherto confined to distinct models.
The unveiling of the SeamlessM4T model resonates with the visionary outlook of Meta’s CEO, Mark Zuckerberg. He envisions this technological innovation as a pivotal instrument in enabling harmonious interactions within the metaverse – a network of interconnected virtual worlds that Meta is ambitiously championing.
This innovative AI offering, available for non-commercial use, is the latest addition to Meta’s series of free AI models introduced this year. One of the notable models, the Llama language model, has posed a formidable challenge to proprietary models backed by tech giants Microsoft and Alphabet’s Google.
In embracing an open AI ecosystem, Meta adopts a strategic approach that capitalizes on the collective wisdom of the global AI community. CEO Zuckerberg asserts that the collective creation of consumer-oriented tools for Meta’s social platforms takes precedence over monetizing access to AI models.
Nevertheless, Meta’s pioneering endeavors are not exempt from legal complexities that pervade the AI domain. Amid this burgeoning landscape, Meta faces legal scrutiny pertaining to the sources of training data that underpin their AI models. In a recent legal development, comedian Sarah Silverman and other authors lodged copyright infringement lawsuits against both Meta and OpenAI, alleging unauthorized utilization of their works as training data.
The foundation of the SeamlessM4T model lies in Meta’s meticulous curation of audio training data derived from a colossal pool of publicly available raw audio content. Although specifics about the audio data repository remain undisclosed, the model’s text data stems from datasets harvested from Wikipedia and its affiliated platforms.
As Meta fortifies its commitment to harnessing AI for transformative applications, it navigates the intricate intersection of AI development, ethical considerations, and legal obligations.