Abstract—We review developments in the SRI Language Modeling Toolkit (SRILM) since 2002, when a previous paper on SRILM was published. These developments include measures to make training from large data sets more efficient, to implement additional language modeling techniques (such as for adaptation and smoothing), and for client/server operation. In addition, the functionality for lattice processing has been greatly expanded. We also highlight several external contributions and notable applications of the toolkit, and assess SRILM’s impact on the research community. I.

This paper investigates the combination of different neural network topologies for probabilistic feature extraction. On one hand, a five-layer neural network used in bottle neck feature extraction allows to obtain arbitrary feature size without dimensionality reduction by transform, independently of the training targets. On the other hand, a hierarchical processing technique is effective and robust over several conditions. Even though the hierarchical and bottle neck processing performs equally well, the combination of both topologies improves the system by 5% relative. Furthermore, the MFCC baseline system is improved by up to 20 % relative. This behaviour could be confirmed on two different tasks. In addition, we analyse the influence of multi-resolution RASTA filtering and long-term spectral features as input for the neural network feature extraction. Index Terms: probabilistic features, bottle neck, hierarchical processing, LVCSR

Abstract—Robust speech recognition under noisy conditions like in human-robot interaction (HRI) in a natural environment often can only be achieved by relying on a headset and restricting the available set of utterances or the set of different speakers. Current automatic speech recognition (ASR) systems are commonly based on finite-state grammars (FSG) or statistical language models like Tri-grams, which achieve good recognition rates but have specific limitations such as a high rate of false positives or insufficient rates for the sentence accuracy. In this paper we present an investigation of comparing different forms of spoken human-robot interaction including a ceiling boundary microphone and microphones of the humanoid robot NAO with a headset. We describe and evaluate an ASR system using a multipass decoder – which combines the advantages of an FSG and a Tri-gram decoder – and show its usefulness in HRI. I.