Conversational Speech Transcription Using Context-Dependent Deep Neural Networks

Frank Seide

Microsoft Research Asia

Tuesday, October 4, 2011
12:30pm - 1:30pm

We apply the recently proposed Context-Dependent Deep-Neural-Network HMMs, or CD-DNN-HMMs, to speech-to-text transcription. For single-pass speaker-independent recognition on the RT03S Fisher portion of phone-call transcription benchmark (Switchboard), the word-error rate is reduced from 27.4 percent, obtained by discriminatively trained Gaussian-mixture HMMs, to 18.5 percent - a 33 percent relative improvement.

CD-DNN-HMMs combine classic artificial-neural-network HMMs with traditional tied-state triphones and deep-belief-network pre-training. They had previously been shown to reduce errors by 16 percent relatively when trained on tens of hours of data using hundreds of tied states. This work takes CD-DNN-HMMs further and applies them to transcription using over 300 hours of training data, over 9000 tied states, and up to 9 hidden layers, and demonstrates how sparseness can be exploited. On four less well-matched transcription tasks, we observe relative error reductions of 22.28 percent.