(NewsNation) — Meta has announced the launch of SeamlessM4T, an AI-powered service that can power translations in text and speech.
SeamlessM4T has speech recognition for 100 languages, Meta announced, and provides four translation options. The tool will support speech-to-text, speech-to-speech, text-to-text and and text-to-speech translation, using the framework of Meta’s Massively Multilingual Tool, which offers speech recognition and identification for more than 1,100 languages.
The number of input and output languages varies for each of the translation modes.
Meta isn’t the only player in the translation game. Google is another big name, with Google Translate offering text-based translations. Google is also working on what it has termed the Universal Speech Model, which is an effort to recognize the world’s 1,000 spoken languages.
As with other investments in artificial intelligence, AI translation services come with potential risks. Automated speech recognition can show biases, with tools less likely to accurately transcribe Black people’s speech when compared to white people’s spoken words.
AI can also replicate other biases, like sexism. That can be especially challenging as some languages are heavily gendered while others are not. Google Translate, for example, has been accused of stereotyping when the tool translates speech about doctors using male pronouns while opting for female pronouns when talking about nurses.