In this submission for MIREX 2013 we utilize an efficient latent variable model for multiple-F0 estimation and note tracking. The model is based on probabilistic latent component analysis and uses pre-extracted note templates from multiple instruments. The templates are also preshifted across log-frequency in order to support pitch deviations and frequency modulations. Contrary to typical shift-invariant models which need to perform convolutions for estimating model parameters, the present model avoids such computations by using the aforementioned pre-shifted templates. Three system variants are submitted: one trained on orchestral instruments for multiple-F0 estimation, one trained on orchestral instruments and piano for note tracking, and a final one trained on piano templates for pianoonly note tracking.