Skip to content

23% More Accurate: AI Model Translates 101 Languages in Real-Time

A new AI model, SeamlessM4T, developed by Meta, can translate speech from 101 languages directly into speech in 36 other languages. This model bypasses the traditional three-step process of speech-to-text, text-to-text, and text-to-speech translation, reducing errors and increasing efficiency. SeamlessM4T achieves 23% higher accuracy in text translation compared to leading models. It uses parallel data mining to learn translations from web data, matching sounds to subtitles. Despite its advancements, human translators remain essential for cultural context and accuracy in critical fields like medicine and law. The model is open-source, encouraging further development, but some experts question its speed and responsiveness compared to alternatives like Google’s model. However, Meta claims a newer version of Seamless is as fast as human interpreters, hinting at the potential for instant interpretation in the future.

Source: www.technologyreview.com

Related Links

Related Videos