In this paper we describe a method to perform sequencediscriminative training of neural network acoustic models without the need for frame-level cross-entropy pre-training. We use the lattice-free version of the maximum mutual information (MMI) criterion: LF-MMI. To make its computation feasible we use a phone n-gram language model, in place of the word language model. To further reduce its space and time complexity we compute the objective function using neural network outputs at one third the standard frame rate. These changes enable us to perform the computation for the forward-backward algorithm on GPUs. Further the reduced output frame-rate also provides a significant speed-up during decoding. We present results on 5 different LVCSR tasks with training data ranging from 100 to 2100 hours. Models trained with LF-MMI provide a relative word error rate reduction of ∼11.5%, over those trained with cross-entropy objective function, and ∼8%, over those trained with cross-entropy and sMBR objective functions. A further reduction of ∼2.5%, relative, can be obtained by fine tuning these models with the word-lattice based sMBR objective function.

No abstract

No abstract

No abstract

No abstract

We summarize the accomplishments of a multi-disciplinary workshop exploring the computational and scientific issues surrounding zero resource (unsupervised) speech technologies and related models of early language acquisition. Centered around the tasks of phonetic and lexical discovery, we consider unified evaluation metrics, present two new approaches for improving speaker independence in the absence of supervision, and evaluate the application of Bayesian word segmentation algorithms to automatic subword unit tokenizations. Finally, we present two strategies for integrating zero resource techniques into supervised settings, demonstrating the potential of unsupervised methods to improve mainstream technologies.

Long Short-Term Memory networks (LSTMs) are a component of many state-of-the-art DNN-based speech recognition systems. Dropout is a popular method to improve generalization in DNN training. In this paper we describe extensive experiments in which we investigated the best way to combine dropout with LSTMs-specifically, projected LSTMs (LSTMP). We investigated various locations in the LSTM to place the dropout (and various combinations of locations), and a variety of dropout schedules. Our optimized recipe gives consistent improvements in WER across a range of datasets, including Switchboard, TED-LIUM and AMI.Projected LSTMs (LSTMPs) [4] are an important component of our baseline system, and to provide context for our explanation of dropout we will repeat the equations for them; here xt is the

We present a new framework for the evaluation of speech representations in zero-resource settings, that extends and complements previous work by Carlin, Jansen and Hermansky [1]. In particular, we replace their Same/Different discrimination task by several Minimal-Pair ABX (MP-ABX) tasks. We explain the analytical advantages of this new framework and apply it to decompose the standard signal processing pipelines for computing PLP and MFC coefficients. This method enables us to confirm and quantify a variety of well-known and not-so-well-known results in a single framework.