ABSTRACTarXiv1610.05256v1[cs.CL]17Oct2016.pdfVIP

ACHIEVING HUMAN PARITY IN CONVERSATIONAL SPEECH RECOGNITION W. Xiong, J. Droppo, X. Huang, F. Seide, M. Seltzer, A. Stolcke, D. Yu and G. Zweig Microsoft Research Technical Report MSR-TR-2016-71 arXiv:1610.05256v1 [cs.CL] 17 Oct 2016 ABSTRACT Conversational speech recognition has served as a ?agship speech recognition task since the release of the DARPA Switchboard corpus in the 1990s. In this paper, we measure the human error rate on the widely used NIST 2000 test set, and ?nd that our latest automated system has reached human parity. The error rate of professional transcriptionists is 5.9% for the Switchboard portion of the data, in which newly acquainted pairs of people discuss an assigned topic, and 11.3% for the CallHome portion where friends and family members have open-ended conversations. In both cases, our automated system establishes a new state-of-the-art, and edges past the human benchmark. This marks the ?rst time that human parity has been reported for conversational speech. The key to our system’s performance is the systematic use of convolutional and LSTM neural networks, combined with a novel spatial smoothing method and lattice-free MMI acoustic training. Index Terms— Conversational speech recognition, convolutional neural networks, recurrent neural networks, VGG, ResNet, LACE, BLSTM, spatial smoothing. 1. INTRODUCTION Recent years have seen human performance levels reached or surpassed in tasks ranging from the games of chess and Go [1, 2] to simple speech recognition tasks like carefully read newspaper speech [3] and rigidly constrained smallvocabulary tasks in noise [4, 5]. In the area of speech recognition, much of the pioneering early work was driven by a series of carefully designed tasks with DARPA-funded datasets publicly released by the LDC and NIST [6]: ?rst simple ones like the “resource management” task [7] with a small vocabulary and carefully controlled grammar; then read speech recognition in the Wall Street Journal task [8]; then Br