Separation of periodic and aperiodic sound components by employing frequency estimation

Separation of a sound to its two main components, periodic and aperiodic ones, is an area that has gained focus in the past years. Instrument classification, speech recognition and many other systems can benefit from such a representation of the digital sound. Unfortunately the separation of the two components is usually not a straightforward process, regarding that the sinusoidal and the noise-like parts of a waveform are not orthogonal, and can easily overlap in frequency. This article presents an efficient and elegant way of separating the two components. The algorithm operates in frequency domain and does the separation on the grounds of the frequency estimation method proposed by Brown. Being simple to implement, in the same time efficient in the task, it can be employed in many current DSP systems.