In spite of their superior performance, neural probabilistic language models (NPLMs) remain far less widely used than n-gram models due to their notoriously long training times, which are measured in weeks even for moderately-sized datasets. Training NPLMs is computationally expensive because they are explicitly normalized, which leads to having to consider all words in the vocabulary when comp...

We present a highly effective and extremely simple noise-robust front end based on novel post-processing of standard MFCC features. It performs remarkably well on the Aurora 2.0 noisydigits database without requiring any increase in model complexity. Compared to the Aurora 2.0 baseline system, our technique improves the average word error rate by 45% in the multi-condition training case, (match...

Recognising speech in the presence of non-stationary noise presents a great challenge. Missing data techniques allow recognition based on a subset of features which reflect the speech and not the interference, but identifying these valid features is difficult. Rather than relying only on low-level signal features to locate the target (such as energy relative to an estimated noise floor), we can...

Stereo-based Piecewise Linear Compensation for Environments (SPLICE) is a general framework for removing distortions from noisy speech cepstra. It contains a non-parametric model for cepstral corruption, which is learned from two channels of training data. We evaluate SPLICE on both the Aurora 2 and 3 tasks. These tasks consist of digit sequences in five European languages. Noise corruption is ...

In this study, the performance of an auditory-model featureextraction “front end” was assessed in an isolated-word speech recognition task using a common hidden Markov model (HMM) “back end”, and compared with the performance of other feature representation front-end methods including mel-frequency cepstral coefficients (MFCC) and two variants (Jand L-) of the relative spectral amplitude (RASTA...

Continuous-valued word embeddings learned by neural language models have recently been shown to capture semantic and syntactic information about words very well, setting performance records on several word similarity tasks. The best results are obtained by learning high-dimensional embeddings from very large quantities of data, which makes scalability of the training method a critical factor. W...