Speech recognition has been around for a while now and has made great strides. Think of Apple’s Siri and Amazon’s Alexa, which are used by millions of people every day. Yet speech recognition still works best when spoken clearly or in a simple or narrow specialised domain. The challenge is to take this even further, so that speech recognition also works when the circumstances are more difficult. Think for example of spoken dialect where the training possibilities of the AI are often limited by the smaller data sets. Or in cases of a speech impediment after a stroke, for example. Another new direction is the detection of emotion in speech and using this to better recognise the context of what is being said.
Within the Radboud University and Radboudumc, together with Telecats, work is being done on more advanced forms of speech recognition in which atypical speech is also better recognised. For this podcast we talk to:
- Henk van den Heuvel (Radboud University CLST)
- Sandra van Dulmen (Radboudumc)
- Arjan van Hessen (Telecats)
The podcast is moderated by Martijn Kriens (Briskr)