Summary

  • A new AI system called Spatial Speech Translation enables headsets to translate multiple speakers’ words into the wearer’s language in real-time.
  • The system works with off-the-shelf noise-cancelling headphones and identifies speakers by dividing the surrounding area into sections and using a neural network to pinpoint the direction from which speech is coming.
  • It then uses public data sets to translate the language and also clones the speaker’s voice tone and characteristics so the translated response sounds like it’s from the original speaker.
  • The technology is a significant improvement on existing systems that only work with single speakers, are not real-time and use automated robotic translation.
  • The next step for researchers is to reduce the time it takes for AI translation to occur following speech.

By Rhiannon Williams

Original Article