文摘
We introduced a set of conversational English speech corpora with high acoustic quality and limited vocabulary derived from the Switchboard-I and Fisher datasets. We investigated numerous state-of-the-art submodular function optimization procedures, including SCSC/SCSK, DS and SFM optimization. We surveyed different submodular function instantiations, where both ‘‘acoustic quality’’ and vocabulary size are adeptly measured via various submodular functions. We provided baseline word recognition results on all of the resultant speech corpora for both Gaussian mixture model (GMM) and deep neural network (DNN)-based systems. We had released all of the corpora definitions and Kaldi training recipes for free in the public domain.