Hybrid sinusoidal modeling of music with near transparent audio quality

It is often believed that sinusoidal as well as sinusoidal plus noise modeling is not capable of delivering high audio quality for complex signals such as wideband music. We identify the key sources of modeling artifacts in sinusoidal modeling systems and demonstrate a hybrid system that offers near transparent quality of reconstructed audio, thanks to application of a dedicated transient model, an accurate parameter estimation method, an advanced tracking algorithm and a warped frequency spectral model of noise.