Fig. 8From: Front-end technologies for robust ASR in reverberant environments—spectral enhancement-based dereverberation and auditory modulation filterbank featuresASR performance of AMFB features with SGMM-HMM for Dev. The SE algorithm, MVN, and a trigram LM are used for evaluation. The thick bars represent the average WERs for SimData and the thin bars for RealData. The average results with GMM-HMM in Fig. 7 are illustrated here for comparisonBack to article page