Efficient spectral envelope estimation from harmonic speech signals

Modern speech synthesis and conversion systems based on statistical methods require vocoders that allow accurate analysis and high-quality reconstruction of speech signals. Reliable spectral envelope estimation and parameterisation is one of the key parts of the vocoding process. Moreover, its efficiency is essential for these systems to be practical in online applications such as speech-to-speech translation or live voice conversion. The simple and fast spectral analysis procedure that is presented in this reported work gives satisfactory results in terms of accuracy and proves useful in vocoding applications.