ENRICH tackles many distinct forms of speech including healthy, disordered, non-native, casual, and computer-generated speech. Improvements in message understanding and cognitive effort due to enrichment have been quantified in ENRICH using physiological measures (pupil size or brain activity measured by EEG), dual-task designs, subjective judgements, identification scores, and task completion rates. Using advanced virtual reality software to create well-controlled but highly-realistic simulations, speech modifications have been tested with normal-hearing and hearing-impaired listeners, non-native listeners and participants with enhanced listening skills.
While synthetic speech has been shown to be less intelligible than natural speech, studies at the University of Edinburgh and the University of the Basque Country examined pupil response measures to computer-generated speech, finding that it is also more effortful to process. Using EEG, researchers at Fraunhofer IDMT found that speech enhancement algorithms reduce correlates of listening effort even when intelligibility cannot be improved further.
Other studies have demonstrated how listening effort varies across different listener groups. In collaboration with Sonova, researchers at UCL measured the extent to which older, hearing-impaired listeners suffer more from fast speech and reverberant speech. Other lister cohorts show enhanced capabilities: a study at the University Medical Centre Groningen showed that musicians outperform non-musicians in tasks involving identifying words from one talker in the presence of a competing talker, in part by attenting more to durational information in speech.
Speakers modify the way they produce speech in challenging conditions, and insights from studying exactly what they do in such scenarios can feed into better algorithms. Scientists at Radboud University Nijmegen collected a large corpus of speech produced in noise by both native and non-native talkers, showing that natural enrichment strategies are common across the two types of talkers and equally beneficial for listeners. Studies at Horzentrum examined the role of visual information, head orientation and gaze changes in complex multi-talker listening tasks, leading to recommendations for future hearing aid strategies.
Scientists from ENRICH have also provided open source software tools that enable measurement of listener preferences for arbitrary speech modifications, a toolkit for pupillometry and a number of new audio and audiovisual speech corpora for other researchers to use.
Results have been disseminated in 60 papers, 48 talks, 66 posters and 21 demonstrations. ENRICH organised a major public understanding event at the UK's Royal Institution (London, 2020), an Industry Event at the International Congress on Acoustics (Aachen, 2019), and a (virtual) Show and Tell event at the International Conference on Acoustics, Speech and Signal Processing (Barcelona, 2020). ENRICH also organised a large-scale international blind evaluation of intelligibility-enhancing speech modifications, the Hurricane 2.0 Challenge, with listener panels in three European countries; results were disseminated at the (virtual) International Conference on Speech Communication, Interspeech (Shanghai, 2020). ENRICH Early Stage Researchers have won Best Paper and Best Poster prizes at several international conferences. A youtube video describing ENRICH has been viewed more than 1100 times.
