|
||
Website deprecated and outdated. Click here for the new site. | ||
Dr.-Ing. Christian HackerAlumnus of the Pattern Recognition Lab of the Friedrich-Alexander-Universität Erlangen-NürnbergAutomatic Assessment of Children Speech to Support Language Learning
Classification of the focus of attention
In the BMBF-Project SmartWeb it is a sub-goal to automatically recognize, whether the user is talking to the system (On-Talk) or to someone else (Off-Talk). This way, no push-to-talk button is required any more. Since the system is beeing developed for a mobile device (T-Mobile MDA pro), we can use the camera of the mobile phone to "look" at the user. With the Viola-Jones algorithm the gaze direction of the user is detected; in the audio signal, prosodic changes of the voice are analyzed.
Recognition of children's speech
At the LME we worked within the EU-project PF-STAR in the research fields 'Speech technologies for children' and 'Technologies for emotions'; the analysis of emotional user states is further focus in the EU network of excellence HUMAINE. For this purpose a corpus with emotional childrens' speech has been recorded (children talking to the AIBO-Robot ). In PF-STAR, English (non-native) and German read speech has been collected from children; it is beeing compared with native speech from English children recorded from the University of Birmingham
Scoring of children's pronunciation (2nd language learners)To automatically assess children's speech, wrong pronounced words are detected by the system and an overall mark of the children's pronunciation is calculated. The automatic scoring is based on more than 100 pronunciation and prosodic features. Different meassures to evaluate the agreement of the automatic score and teachers' marks are evaluated (in cooperation with the OHM-Gymnasium , Erlangen). CALLER (Computer Assisted Language Learning from ERlangen) is a client/server application: The program running in a browser can be used by children to exercise English (diploma thesis A. Hessler), while their pronunciation is analyzed automatically on the server. |