[国内学会] Efficient Pitch Estimation on Natural Opera-Singing by a Spectral Correlation based Strategy

情報処理学会 第107回音声言語情報処理研究発表会

We present in this work a study for robust pitch estimation on signals presenting wide-range pitch content, as is the case of opera singing. Aiming to perform automatic features extraction for the further development of parametric opera singing synthesis technology we evaluate four state-of-the-art pitch estimators, reporting in particular technical details of one introduced in previous work, based in a technique called Spectral Amplitude Autocorrelation (SAC). The results issued from subjective and objetive evaluations show clear performance trends, denoting robust estimation performance for SAC without observing significant sensitivity to the pitch height.