No abstract

No abstract

Recent work has shown success in using neural network language models (NNLMs) as features in MT systems.Here, we present a novel formulation for a neural network joint model (NNJM), which augments the NNLM with a source context window. Our model is purely lexicalized and can be integrated into any MT decoder. We also present several variations of the NNJM which provide significant additive improvements.Although the model is quite simple, it yields strong empirical results. On the NIST OpenMT12 Arabic-English condition, the NNJM features produce a gain of +3.0 BLEU on top of a powerful, featurerich baseline which already includes a target-only NNLM. The NNJM features also produce a gain of +6.3 BLEU on top of a simpler baseline equivalent to Chiang's (2007) original Hiero implementation.Additionally, we describe two novel techniques for overcoming the historically high cost of using NNLM-style models in MT decoding. These techniques speed up NNJM computation by a factor of 10,000x, making the model as fast as a standard back-off LM.

In this work we formulate a novel approach to estimating the parameters of continuous density HMMs for speaker-independent (SI) continuous speech recognition. It is motivated by the fact that variability in SI acoustic models is attributed to both phonetic variation and variation among the speakers of the training population, that is independent of the information content of the speech signal. These two variation sources are decoupled and the proposed method jointly annihilates the inter-speaker variation and estimates the HMM parameters of the SI acoustic models.We compare the proposed training algorithm to the common SI training paradigm within the context of supervised adaptation. We show that the proposed acoustic models are more efficiently adapted to the test speakers, thus achieving significant overall word error rate reductions of 19% and 25% for 20K and 05K vocabulary tasks respectively.

No abstract