We (and others) have shown that
..it’s possible to recognise disordered speech..
..provided that you can adapt the recogniser to the speaker..
..adaptation needs training data, but typically only a few minutes of speech..
.. it’s best to close the ‘virtuous circle’ where data collected from a device in use is used to improve the performance of the device.
CloudCAST aims to translate this research into a resource for speech professionals.
- See Exemplar Applications for details of what CloudCAST will provide
- See ASR for Disordered Speech for our methodology