Meta launches SeamlessM4T, an AI model to translate speech-to-text, speech-to-speech & more
The SeamlessM4T model has the capability to carry out a range of functions, including converting speech to text, translating speech to speech, translating text to text, and recognising speech.
artificial intelligence
Highlights
- SeamlessM4T is the first integrated multilingual, multimodal AI transcription & translation model
- The model can translate speech-to-text, speech-to-speech, text-to-speech, and text-to-text
- It can cater to speech recognition in nearly 100 languages
In a testament to its commitment to harnessing the power of artificial intelligence (AI), Meta has yet again pushed the boundaries of technological innovation. Having successfully incorporated AI into numerous aspects of its models, the company has now introduced a remarkable multilingual AI translation system.
On 23 August 2023, Meta introduced an advanced AI model called 'SeamlessM4T' designed to revolutionise translation and transcription tasks. This new model can perform a variety of functions like turning speech into text, translating speech, converting text to speech, and translating text.
Supporting a remarkable 100 languages, 'SeamlessM4T' serves as a powerful tool for speech recognition, translation, and synthesis to meet diverse linguistic needs.
A milestone in AI advancement
With the introduction of 'SeamlessM4T,' Meta's dedication to facilitating cross-language communication has taken a huge step forward. This ground-breaking technique is supported by the large 'SeamlessAlign' dataset, which contains 270,000 hours of voice and text alignment. The dataset's publication establishes a significant precedent in the AI sphere, underscoring Meta's commitment to pushing the frontiers of technical innovation.
NLLB, a text-to-text translation model
The release of 'SeamlessM4T' represents the conclusion of Meta's earlier initiatives that emphasised multilingual capabilities. Notably, the company debuted 'No Language Left Behind (NLLB)' last year, a text-to-text translation paradigm that supports an amazing 200 languages.
Furthermore, Meta showcased its Universal Speech Translator, a pioneering speech-to-speech translation system designed for languages that lack commonly employed writing systems, like Hokkien. These initiatives helped pave the way for ‘SeamlessM4T,’ which builds on research from other projects to offer a thorough multilingual and multimodal translation experience.
Towards a globally connected future
With the release of SeamlessM4T, Meta's continued commitment to overcoming linguistic barriers advances much further. Meta seeks to build a future where language barriers do not prevent people from understanding one another by enabling seamless communication across many languages.
COMMENTS 0