| Summary: | 碩士 === 國立臺灣科技大學 === 電機工程系 === 92 === Pitch period is a key feature in speech compression, synthesis, language identification and speech recognitions. With better pitch estimation result in better accuracy. The AMDF algorithm is the most efficient and easiest way of estimating pitch contour in the time domain. However, Due to the falling trend of AMDF curves, it is easy to estimate pitch in doubled or halved. To avoid these errors, several algorithms based on AMDF are discussed and compared. These algorithms need to increase its complexity or to add some useful skills to improve the accuracy of the estimated pitch information. Here an improved-AMDF algorithm named Cross-AMDF is developed and proposed. Finally, its application for voiced / unvoiced segment is also presented.
|
|---|
