| Résumé |
Reliable fundamental frequency (f0) estimation of monophonic sounds has existed for years. However, most of the existing algorithms require manual adjustment of input parameters for optimal performance on a given sound file. Additionally, though near-perfect accuracy has been attained for relatively clean sound files, performance is known to decline drastically for cases with background percussion, noise, or reverberation. The first goal of this work was thus to improve upon the current state of monophonic f0 estimation, not by modifying the estimation algorithm itself, but instead by increasing its automaticity and robustness. By choosing optimal fundamental frequency estimation input parameters for a wide range of files or by setting them algorithmically in relation to the content of each file, the need for user interaction could be eliminated. Additionally, by using an eclectic and challenging database of sound files including many of which had been the source of com- plaint for previous users of f0 estimation software, one could hope to improve performance in more complicated cases. |
