語音辨識是人工智慧相當關注的領域，但受限於不同環境的影響，至今依舊 難有一個系統能如人類般清晰的識別。本研究旨在探討梅爾頻率倒譜系數(MFCCs) 及連接性音頻分類(CTC)在語音辨識系統上的功能性。 本研究使用github 上所提供的無噪聲語料，以不同的處理方式建構遞歸神 經網絡模型，並選定一些變因做為探討比較的對象。;Speech recognition is part of the artificial intelligence that is highly concerned, but is limited by different environmental influences. It is still a difficult subject to have a system that can be clearly identified as humans. This study aims to investigate the functionality of the Mel Frequency Cepstral Coefficients (MFCCs) and the Connectionist Temporal Classification (CTC) on speech recognition systems. This study uses the noise-free corpus provided on github to construct a recursive neural network model in different ways, and selects some variables as the object of discussion and comparison.
