“Capstan is based on an evaluation of the musical content”
Melodyne inventor Peter Neubäcker on the development of Capstan
Peter, what gave you the idea of developing Capstan?
Sound engineers from msm studios in Munich – one of the largest restoration studios in Germany – were enquiring whether there was any way of using Melodyne to eliminate wow and flutter. They were restoring at the time an old Beethoven recording that droned badly... With Melodyne, however, it can't be done, as the rigid coupling of pitch and time is precisely what Melodyne overcomes. “But perhaps there's another way of doing it,” I thought to myself.
Are the Melodyne algorithm and Capstan somehow related, then?
At the time of the Beethoven enquiry, I was engaged in the further development of our DNA algorithm to improve the tracking of pitch in polyphonic material such as multi-part vocals with lots of vibrato. The idea naturally suggested itself, therefore, of applying the newly developed algorithm to wow and flutter. And, lo and behold: in principle it worked a treat! The basis of the detection algorithms of Capstan and Melodyne – a future version of Melodyne, at least – are therefore very similar. Based on this technology we optimized Capstan for special problems with wow and flutter.
 | |
 |
Now the real problem until now was the precise determination of the wow and flutter, as the actual correction process is astoundingly simple: once we've found the wow curve, we just read back the signal faster or slower accordingly. The playback has nothing, then, to do with pitch-shifting or the Melodyne algorithm. It is a simple varispeed process entirely free from artifacts.
For which problems is Capstan suitable?
It doesn't matter much to Capstan how the pitch problem came about. Capstan detects not only periodic fluctuations, due, for example, to a poorly centered hole in a shellac disk, but also chaotic ones, such as those attributable to a sticking tape. In addition to the usual wow and flutter problems, gradual pitch deviations are also detected and corrected.
Is Capstan's effectiveness in any way circumscribed?
Capstan is based on an evaluation of the musical content, regardless of whether it's classical, rock or film music. So already one limitation is clear: wow and flutter in sound effects or speech cannot be treated with Capstan. Of course, these signals are also far less susceptible to them than music, where even the most minute wobbling in pitch of sustained notes can render a recording unusable. But naturally there are also tape defects (not to mention dropouts) so severe that they cannot be remedied with even the finest tools.
With our demo version, everyone can try it out and discover for themselves whether, and how well, Capstan is capable of detecting and eliminating pitch fluctuations.
