INTERSPEECH 2004 - ICSLP
This paper reports a cough detection technique in a spoken dialogue system for health care task. In conventional speech recognition systems, their targets are verbal sounds only and non-verbal sounds like coughs are processed as burden noises which cause misrecognition. Coughing, however, is one of the most important barometers of daily health check, so the cough detection can be useful for checking a progress state and degree of a disease. It therefore is important to recognize not only user's utterances but also user's coughs in the spoken dialogue system for health care. In this paper, based on acoustic analysis results, a cough detection system with two different cough acoustic models constructed from band-pass filtered signals and unfiltered signals is proposed. The experimental results show that this method can improve an accuracy of the cough detection compared with the method which uses one acoustic model.
Bibliographic reference. Takahashi, Shin-ya / Morimoto, Tsuyoshi / Maeda, Sakashi / Tsuruta, Naoyuki (2004): "Cough detection in spoken dialogue system for home health care", In INTERSPEECH-2004, 1865-1868.