using an algorithm that analyzes and decodes the human voice (called vocoder) a group of neuroingenieros of Columbia University has developed a system that promises to co invest the human thoughts in understandable words. When people talk (or imagine speaking) and when they hear others speak, certain patterns of activity are in the brain. Recognize, record, and decode these patterns was the key to the study carried out the equipment of the Dr. Nima Mesgarani, whose main objective is to assist individuals suffering from some kind of disease that prevents them from speaking.» Our voices help us to connect with our friends, family and the world us around, so you lose the power of the voice by a wound or disease is devastating,»said Mesgarani. «With the study today, we have the potential to return that power. We have shown that with the right technology, the thoughts of these people can be decoded and understood by any listener.» The researchers used technology similar to that found in artificial as Alexa or Siri intelligences, virtual attendees at Amazon and Apple. The vocoder is an algorithm which, after being loaded with thousands of recordings of people talking, can recognize patterns of the brain become spectrograms (Visual representations of sound frequencies) and play them as a voice.
The process of encoding and decoding | Image: Nature to assist the vocoder Mesgarani team collaborated with the neurosurgeon Ashesh Dinesh Mehta (one of the co-authors of the study), who specializes in the treatment of patients with epilepsy – some of which tend to submit regularly to surgeries.» Working with Dr. Mehta, we ask patients with epilepsy who were undergoing brain surgery that they heard phrases of different people while we measured the brain activity,»says the leader of the research. «These neural patterns were used to train the vocoder». Then you were asked to those patients that they recite numbers between zero and nine while also employers registered to decode them. The sound was then analyzed by neural networks for the best possible quality, and recited by an artificial intelligence.
«Would give you to any that speech he has lost the chance to reconnect with the world,» said the co-author of the study.
According to the study, those who heard those sounds recognized numbers that had been recited in almost 75 per cent of the cases. «The sensitivity of the vocoder and the power of neural networks represented the sounds that the patients had heard originally with surprising accuracy,» said Mesgarani.El next step is to try to decode more complex words and whole sentences, and the team also will carry out the experiment using patterns that are generated when the individual imagines the words but does not say them out loud. «They hope to be able to use the system as an implant that it converts user thoughts into words in the future.» This could be revolutionary. «Would give you to any lost speech, whether it is a wound, or a disease, the chance to reconnect with the world,» concluded Dr. In this note: