Automatic Cloning of Recorded Sounds by Software Synthesizers

Any audio recording can be turned into a digital musical instrument by feeding it into an audio sampler. However, it is difficult to edit such a sound in musical terms or even to control it in real time with musical expression. This does not change much if a more sophisticated resynthesis method is applied. Many electronic musicians appreciate the direct and clear access to sound parameters a traditional analog synthesizer offers. Can one automatically generate a synthesizer setting that approximates a given audio recording and thus clone a given sound to be controlled with the standard functions of the particular synthesizer employed? Even though this problem seems highly complex, we demonstrate that its solution becomes feasible with today’s computer systems. We compare sounds on the basis of acoustic features known from Music Information Retrieval and apply a specialized optimization strategy to adjust the settings of VST instruments, which is sped up using multi-core processors and networked computers.