Speech Analytics for the Detection of Neurological Conditions in Global English
Sam Hollands
University of Sheffield
Dementia is an umbrella term for the loss of cognitive and memory abilities caused by a wide variety of neurological conditions. It has been discovered that both the content of an individual's discourse and the acoustics of their produced speech can be automatically analysed to help detect dementia and other neurological conditions. Whilst the cutting edge demonstrates effective diagnostic capabilities on L1 (native) speakers of English, this talk will explore ongoing research assessing the efficacy and exploring solutions for L2+ (non-native) English performance. This research treats a dementia classification pipeline as a modular system containing an automatic speech recognition (ASR) component to extract transcribed language; and then the challenge of classifying using features extracted from the acoustic signal and transcribed output. Limitations of ASR across a wide range of L2+ backgrounds will be explored challenging existing beliefs about the competency of state-of-the-art cloud-based ASR APIs on non-native speech and critically assessing the limitations of word error rate (WER) as the ubiquitous metric for ASR evaluation. My talk will then explore ongoing research into the features of dementia, potential issues in the generalisability of sparse dementia corpora, and early work looking at the impact of features of non-native speech.