Restoration of missing data in Greek folk music by interpolation techniques

Music recordings often suffer from noise. The noisy segments may be treated as missing data. To restore them, one may employ interpolation techniques. A music signal is modeled as an autoregressive process, and three interpolation methods are developed that are based on maximum likelihood, Gibbs sampling, and Expectation Maximization. The aforementioned techniques are tested for restoration of missing data in vocal and instrumental Greek folk songs. Experimental results show that interpolation techniques based on maximum likelihood and Gibbs sampling offer better restoration results than Expectation Maximization.