Fig. 2From: Feature compensation based on independent noise estimation for robust speech recognitionAverage log-likelihoods of the clean, combined, and adapted GMMsBack to article page