CAJ | 학술논문

本文提出一种语音激活检测的改进算法。首先在传统噪声估计的基础上，用Bark子带代替了DFT频域变换，目的在于降低计算复杂度；其次将估计的噪声谱进行白化滤波，并运用于子带谱熵算法中的谱熵计算中。把谱熵值作为VAD算法提取的特征参数，通过门限设定与比较，得出最初的VAD判决结果，增加拖尾延迟保护机制得出最终的VAD判决结果。
본문제출일충어음격활검측적개진산법。수선재전통조성고계적기출상，용Bark자대대체료DFT빈역변환，목적재우강저계산복잡도；기차장고계적조성보진행백화려파，병운용우자대보적산법중적보적계산중。파보적치작위VAD산법제취적특정삼수，통과문한설정여비교，득출최초적VAD판결결과，증가타미연지보호궤제득출최종적VAD판결결과。
This paper puts forward an improved al-gorithm of voice activity detection. At first, on the basis of the traditional noise estimation, 129 DFT bins are replaced by 18 Bark scaled frequency bands to reduce the computational complexity. Secondly, bleaching filtering is done to the estimated noise spectrum, and sub band spectrum entropy algorithm is applied to spectrum entropy calculation. Then with the spectral entropy as an extracted feature parame-ters of VAD algorithm, through the threshold setting and comparison, the initial VAD decision is ac-quired. At last, the trailing delay protection mecha-nism is introduced to get the final VAD decision re-sults.