SpeDial Highlights

In SpeDial, industry and academia collaborated seamlessly towards transferring state-of-the-art features and algorithms into commercial systems. Working with real data and services was essential in guiding our solution that has been fully integrated in commercial platforms.

  1. Various features were exploited for the detection of problematic parts (hotspots) of dialogues. High classification accuracy (above 80%) was achieved for different domains and languages. Furthermore, two approaches were investigated for the analysis of the root causes of hotspots.
  2. A rich set of features were developed for a number of major aspects of call-flow and discourse analysis, namely, affective analysis of dialogues, detection of age and gender, intent recognition, detection of disfluency and hyper-articulation. High performance was achieved for all those areas.
  3. Last but not least, the analytics that were integrated in the platforms of Voice Web and Nu Echo were found to reduce the time spent for the analysis and tuning of Spoken Dialogue Systems.

Overall, the SpeDial analytics achieved over 80% accuracy in hotspot detection and 70% for the classification of the respective root causes.
Comments