Automated voice-based testing as a potential recruitment tool in clinical trials

The automation of voice-based cognitive tests using automatic speech recognition (ASR) would open up new recruitment channels into clinical trials. In addition to cognitive test performance, machine learning can be used to exploit the acoustic properties of voice, to infer aspects of health and wellbeing. This requires access to raw audio data. This study examined the possibility of automating voice-based cognitive assessments in two experiments.

The first experiment made use of the Amazon Echo, a consumer-grade system, designed primarily to support simple voice interactions within a home setting. This study aimed to establish feasibility and validity of testing with ASR. The study found that (i) there was excellent correspondence between automated and manual scoring of participant performance, and (ii) automatic speech recognition errors had relatively little impact on scoring performance.

The second experiment extended the first with a digit span task on a device-agnostic web-based system. The aim of the second study was to use ASR with customised interactions, greater control over data storage, and raw audio data. The second study again found (i) there was excellent correspondence between manual and automated scoring, and (ii) weighting end-of-utterance detection made the web-based system less vulnerable to recognition errors.

Cambridge Cognition are implementing an online ASR-based verbal cognitive testing platform, which has operational advantages in a recruitment setting and enables acquisition of raw audio data for voice feature analysis. 

View poster 

Find out more

Tags : voice recognition | automated | voice-based assessment | automatic speech recognition | machine learning