This paper investigates error concealment based on sinusoidal analysis and synthesis. Major shortcomings are identified with focus on the extraction of sinusoidal frequency evolution and sinusoid matching, and a new approach to frame loss concealment is proposed. It involves parallel Fourier transformation with long and short windows to accurately extract model parameters, and is complemented with two sinusoid matching techniques -- sinusoidal pair alignment by dynamic programming and harmonics-based matching. Moreover, due to the incompatibility of sinusoidal representation with broadband, noise-like signals, an alternative, ``sinusoids plus residual'' model is incorporated. The new algorithm was applied to CD quality audio of various genres and was demonstrated to improve the perceptual quality with considerable gains for non-transient frames.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.