本論文以數位信號處理晶片TMS320C6711,發展語音信號即時辨識系統,經由求取欲分析信號之特徵參數,並應用類神經演算法則,發展快速之信號辨識技術,以達成語音信號即時辨識之要求。 本論文的主題包括二大部分:第一部份是利用小波封包分解法將原始訊號分解,再以多重解析為基礎之小波封包為基底,使用希爾伯特轉換將信號變化依時間分解成多個基底函數之分量,進而再轉換成局部之能量頻譜以獲得各語音聲源訊號具代表性的樣板特徵參數;第二部份是利用各樣本類別的特徵參數,經由類神經網路加以訓練,進而執行自動辨識之工作。最後對此二部分整合,並以實測數據驗證研究成果之有效性,整個系統嵌入於數位信號處理晶片TMS320C6711之內,利用數位信號處理器之高度執行效率,建立一套實用的信號數位處理系統,以達到操作簡便,快速而正確的語音信號處理結果。
The Research is using digital signal processing chip TMS320C6711 to develop real time speech recognition. The fast signal identification result is achieved by means of the extraction of feature parameters of the signal and the establishment of a neural network. There are two major topics about this research. The first part is using wavelet packed method to decompose original signal. Thus, use multi-resolution space to be wavelet packet base, and then apply the Hilbert transform can get representative pattern feature parameters of each sample classification individually. The second part is using each of feature parameters to train in neural network, and then the automatic identification task is implemented. Finally, the two parts are integrated to perform real-time speech recognition system. An evaluation will execute to demonstrate the effective performance of this system. The system is embedded in a digital signal processing chip TMS320C6711 take advantage of the high efficiency of performance of the DSP, an easy use, fast and accuracy real-time DSP based speech signal processor system will be established.