Meta has recently unveiled a new artificial intelligence (AI) model that can translate and transcribe nearly 100 languages, with the aim of pioneering a universal translator. This single-system model, named SeamlessM4T, is designed reduce errors and delays commonly associated with translation processes.
At the same time, Meta has also released the SeamlessAlign open multimodal translation dataset, which contains 270,000 hours of speech and text alignments.
Mark Zuckerberg, CEO of Meta, has long expressed a desire to use AI to facilitate interaction between people in the company’s envisioned ‘metaverse’, a set of interlinked virtual worlds. The SeamlessM4T model is being made available for non-commercial use, paving the way for a more open AI ecosystem, according to a blog post from the company. “Over time, we will integrate these advances in translation and transcription through artificial intelligence into Facebook, Instagram, WhatsApp, Messenger and Threads,” he said.
The release is part of a flurry of AI models launched by the world’s largest social media network this year. One of these is a large language model called Llama, which poses a major challenge to expensive proprietary models from tech juggernauts like Microsoft-backed OpenAI and Alphabet’s Google.
By combining speech recognition and translation advancements with AI resources, Meta is hoping to make global conversations easier and more accessible. In the near future, the company’s AI model may very well do away with language barriers forever.
1 thought on “New AI Model Developed by Meta to Create Universal Translator”