Construction of a Corpus for Elderly Japanese Speech Recognition

We have constructed a new speech data corpus using the utterances of 144 elderly Japanese people (average age: 81.1), to improve the accuracy of automatic recognition of the speech of old people. For elderly, it is proposed that daily verbal interaction with others help to prevent or restrain progress of dementia. Therefore, developing the nursing robot with sophisticated spoken dialogue system has been expected. However, speech recognition for elderly using standard recognition model is not able to achieve sufficient performance. It is assumed that making the acoustic model of elderly is essential to improve speech recognition for the elderly. Thus, we are continuing to collect more utterances in various parts of Japan, and attempt to construct a new corpus suitable for elderly acoustic features.