本論文旨在建構一套盲源分離軟體，並導入深度學習演算法，以此實現音頻擷取時的降噪功能，並於未來應用於智慧製造的領域。 實驗流程方面，本論文先以乾淨的音源訊號進行模型訓練，以此驗證本論文建構之軟體的正確性，再將各式噪音與原始訊號合成，研究應用深 度學習於音頻降噪實現時最佳的模型訓練方式。;The propose of the thesis is to build a blind source separation program which based on deep learning algorithm. The goal is to achieve noise reduction while capturing sound signals, and to applicate in intelligent manufacturing in the future. The process of experimentation is training the model with clean audio signal at first. It would verify if the program be written properly. After the verification, the original signal combined with noise would be used to train the model for researching the best training method on noise reduction.
