Research on speech recognition interface for elderly people based on acoustic feature extraction of elderly speech

本研究旨在研究老年语音的声学特征，以提高老年说话人的语音识别率。听力测试是为了了解老年人和非老年人言语的主观印象的差异。我们总结了老年人言语的特点是三个形容词短语：声音沙哑、声音不清和说话缓慢。然后研究每个形容词短语对应的老年语音的物理特征。结果表明，老年语音的声学特征是：平均功率谱在高频区能量升高，语音平稳时频谱起伏，相邻音素间转换速度慢，发音速度慢，发声节奏不规律。

This research aims at investigating acoustic features of elderly speech for improving speech recognition rate of elderly speakers. Listening tests are carried out to know difference of subjective impressions between elderly and non-elderly speech. We summarized that elderly speech was characterized by three adjective phrases : hoarse voice, ill-articulated voice and slow utterance. Then a physical characteristic of elderly speech is investigated corresponding to each adjective phrase. As the result, acoustic characteristics of elderly speech are ; energy lift in higher frequency region in average power spectra, fluctuations in spectra during stationary parts of utterance, slow transition speed between adjacent phonemes, slow utterance speed and irregular utterance rhythm.