Skip to main content

Table 2 Average SimData and RealData word error rates for the REVERB Challenge development set

From: Feature enhancement of reverberant speech by distribution matching and non-negative matrix factorization

  1. The following abbreviations are used in the table: multicondition (MC), baseline (BL), speaker adaptation (SA), linear discriminant analysis (LDA), maximum likelihood linear transform (MLLT), distribution matching (DM), missing data imputation (MDI), non-negative matrix factorization (NMF), speaker adaptive training (SAT), delay-and-sum (DS), feature domain boosted maximum mutual information criterion (f-bMMI), deep neural network (DNN), minimum Bayes risk (MBR) decoding and discriminative training with state-level minimum Bayes risk (SMBR) criterion. The dashed line separates the performance-wise comparable back-ends