The traditional approach to assessing spoken English is to have a well-trained human assessor listen to the test - either live or recorded - and mark the performance on a standardised scale. There are two main problems with this. First the process is highly expensive as it requires the training of an assessor. Secondly the process is not scaleable to large numbers of candidates. There is considerable interest in automating this approach to address these problems. The goal of this project is to develop techniques to automatically evaluate oral communication skills in collaboration with Cambridge University ESOL. The project will make use of state-of-the-art automatic speech recognition (ASR) approaches to provide transcriptions and features that characterise the communications skills of the candidate.

The figure above describes an overview of the approaches that will be taken.

Specific areas that may be examined include:

• adapting a speech recognition system to non-native speakers;
• automatic correction of crowd-sourced transcriptions;
• use of 'crowd-sourced' transcriptions to train speech recognition systems;
• extracting features from transcriptions (either from the ASR system or crowd-sourced) for assessing English;
• designing a classifier given a set of features for spoken English assessment as a second language.

There is funding available on this project for short-term contracts or studentships at Cambridge University. If you are interested please contact Prof Mark Gales.