Speech-to-text transcription for accented speech:

Brisbane Convention & Exhibition Centre 

Speech to text apps work very well in most situations. For this to work, three key ingredients are a) clear speech, b) audibility and c) connectivity. However, accuracy can decrease with accented speech stemming from hearing loss or with English as a second language. To demonstrate these factors in the context of speech-to-text apps, we present results that compare human and automated transcription of 60+ talks over 5 days at a recent scientific conference. At the same time, the presentation will make use of Google’s Project Relate, which is a speech to text app customized for accented speech. On a related note, Google’s Project Euphonia is targeted at patients with cerebral palsy, Parkinson’s or Amyotrophic Lateral Sclerosis (ALS) which could open up new opportunities for functional neuroimaging.