Meta announced the release of an artificial intelligence model named SeamlessM4T
Meta announced the release of an artificial intelligence model named SeamlessM4T, endowed with the capability to identify nearly 100 languages. This versatile model not only supports text-to-text and voice-to-text translations but can also transmute voices from approximately 100 languages into auditory outputs in 36 distinct languages, including English, or render them into written form in 35 languages, with English among them.
SeamlessM4T has undergone rigorous training, leveraging approximately 270,000 hours of corresponding voice and textual data. In comparison to other linguistic translation tools that encompass only a fraction of global languages, many tend to falter with glaring inaccuracies or vague translations when interpreting languages from different cultural backgrounds, often diminishing the overall translation efficiency. Meta asserts that by integrating multiple language translation models, SeamlessM4T will facilitate translations across multiple languages in a more efficient and accurate manner.
Launching SeamlessM4T, Meta’s CEO, Mark Zuckerberg, perceives it as a significant stride toward enhancing communication amongst speakers of diverse languages, especially pivotal in virtual worlds such as the Metaverse, where direct language interaction remains a crucial mode of engagement. Furthermore, Meta intends to offer the SeamlessM4T model for non-commercial purposes to the wider community.
Beyond Meta, industry giants including Google, Microsoft, and Amazon persistently refine their translation tools. Meanwhile, entities like OpenAI are increasingly harnessing artificial intelligence to amplify their natural language understanding and translation capabilities, thus enabling users to interact seamlessly with AI through natural language.