哭泣為嬰兒最初與外界交流的語言，亦為主要表達其需求與情緒之溝通手法，透過發出哭聲訊號引起照顧者之回應，以滿足其需求。不同類型的哭聲模式之間存有差異，因此哭聲成為辨識嬰兒不同需求與狀態之重要訊息來源。然而，新手父母或缺乏經驗的照護人員難以單憑哭聲即能了解嬰兒哭泣之原因，嬰兒哭泣亦成為照顧嬰兒時需面對的主要問題。嬰兒哭泣原因不計其數，根據依附理論可得知嬰兒會因缺乏安全感而有尋求安全感之需求，但尚未有研究辨識此哭聲類型，因此，相較其它研究之哭聲辨識模型常見的哭泣原因外，本研究還多考慮缺乏安全感之哭聲類型。 本研究旨在建立哭聲自動判斷模型並採用真實哭聲數據庫，研究流程使用梅爾倒頻譜係數(Mel-Frequency Cepstral Coefficients, MFCC)作為特徵提取，並以長短期記憶網絡(Long Short Term Memory networks，LSTM)作為分類模型，最終分為五種哭聲類型，即肚子餓、想睡覺、大小便、生氣與缺乏安全感。實驗結果顯示模型 之 精度 (Precision)為 48% 有 鑑於此，嬰兒哭泣問題有望自動解決。;Crying is the first language of an infant to communicate with the external world. It’s also the primary means of communication that expresses infant’s needs and emotions. Infant crying is a signal that elicits caregivers to respond to meet his/her needs. There are variations between different types of crying patterns. Therefore, crying becomes an important source of information to distinguish different needs and conditions of infants. However, it is challenging for novice parents or inexperienced caregivers to understand the reason of infant crying. Infant crying is also the main trouble when caring for infants. According to the attachment theory, it can be known that infants have a need to seek security from the main caregivers due to lack of security. But there is no research to recognize this type of crying. Consequently, compared with other studies, this study also considered the cause of insecurity. This study aims to construct an automatic infant cry recognition model and use a real crying database. The acoustic features are extracted using Mel-Frequency Cepstral Coefficients (MFCC), and Long Short Term Memory networks (LSTM) is used to classify crying signal into five categories including want to sleep, poopee, anger and insecurity. The experimental results show that the model achieves 48% precision, indicating the infant crying problem is expected to be resolved automatically.
