The source code and examples are related to our work on bounded conditional mean imputation (BCMI). We have studied truncated posterior mean imputation1 (TPMI) and truncated conditional mean imputation2 (TCMI). TPMI and TCMI substitute unreliable feature components with GMM-MMSE estimates. For reference, we include cluster-based imputation3 that substitutes unreliable feature components with GMM-MAP estimates.
Reference: U. Remes, A. Ramírez López, K. Palomäki, and M. Kurimo, Bounded conditional mean imputation with observation uncertainties and acoustic model adaptation, IEEE/ACM TASLP, 23(7), pp. 1198-1208, 2015.
For more information, contact: Ulpu Remes, Ana Ramírez López, Kalle Palomäki, Mikko Kurimo
DOWNLOAD (2.2 MB) includes reconstruction and mask estimation functions with a demo.
Main functions: