The Challenge (which will be managed via our Kaggle competition) will require contestants to develop a solution for automatically scoring phone-recorded responses to a memory test – as part of the Valenzuela lab’s LOGOS dementia diagnosis project.
Speech recognition in this setting is actually more difficult than transcribing natural speech, as the words are random and hence there is no semantic context to help algorithms guess what was just said. In fact, Google’s transcription models perform only about 65% accurately on this task.
Think you can do better? Register for the Challenge using the form below!
MAJOR HACK! MAJOR HACK!