April 25, 2019

expert reaction to study decoding brain activity to speech

Research published in Nature demonstrates a neural decoder that uses cortical activity to synthesize audible speech.

Prof Tom Otis, Chief Scientific Officer at Sainsbury Wellcome Centre for Neural Circuits, University College London (UCL), said:

“This is a foundational paper which demonstrates the feasibility of a brain prosthetic for speech production. It is an important step forward in research for patients suffering from loss of speech as a result of various neurological insults such as neurodegenerative diseases or injuries to the brainstem, spinal cord and vocal musculature.

“Although the study was limited to 5 patients without medical conditions affecting their speech, the performance of the processing algorithm converting neural signals into synthesised speech was impressive.

“Such early success is quite encouraging as the algorithm will certainly improve and there are more refined methods for recording neural signals that could yield better performance of an eventual medical device.

“Based on observations of patients fitted with other types of brain machine interface prosthetics such as prosthetic limbs and cochlear implants, it is also likely that in the future, should this technology come to clinical practice, patients implanted with a speech production prosthetic would undergo a phase of learning as they used the device. This would result in improvements in the intelligibility of the speech produced over time.”

Prof Sophie Scott, Group Leader for Speech Communication Neuroscience Group, University College London (UCL), said:

“This is very interesting work from a great lab but it must be noted that it is at very early stages and is not close to clinical applications yet. This work asked listeners to try and recognise the speech produced through this technique but they were able to select from a closed set of options – a list of 25 or 50 words. This makes the task easier and likely increases recognition rates as they are choosing from a small selection. Compare that to the 12,000 words a 12 year old human know or the 23,000 words an adult does and you can see that there is some way to go to having full real-world relevance as of yet. It will be interesting to follow the progress of this work.”

‘Speech synthesis from neural decoding of spoken sentences’ by Gopala Anumanchipalli et al. was published in Nature at 18:00 UK time on Wednesday 24th April.

Declared interests

Prof Tom Otis: no conflicts of interest.

None others received.