Fig. 3From: Towards end-to-end speech recognition with transfer learningHigh-level features extraction using CNMFBack to article page