Improving LVCSR system combination using neural network language model cross adaptation

State-of-the-art large vocabulary continuous speech recognition (LVCSR) systems often combine outputs from multiple subsystems developed at different sites. Cross system adaptation can be used as an alternative to direct hypothesis level combination schemes such as ROVER. The standard approach involves only cross adapting acoustic models. To fully exploit the complimentary features among sub-systems, language model (LM) cross adaptation techniques can be used. Previous research on multi-level n-gram LM cross adaptation is extended to further include the cross adaptation of neural network LMs in this paper. Using this improved LM cross adaptation framework, significant error rate gains of 4.0%-7.1% relative were obtained over acoustic model only cross adaptation when combining a range of Chinese LVCSR sub-systems used in the 2010 and 2011 DARPA GALE evaluations. Copyright © 2011 ISCA.

Integrated online speaker clustering and adaptation

For many applications, it is necessary to produce speech transcriptions in a causal fashion. To produce high quality transcripts, speaker adaptation is often used. This requires online speaker clustering and incremental adaptation techniques to be developed. This paper presents an integrated approach to online speaker clustering and adaptation which allows efficient clustering of speakers using the same accumulated statistics that are normally used for adaptation. Using a consistent criterion for both clustering and adaptation should yield gains for both stages. The proposed approach is evaluated on a meetings transcription task using audio from multiple distant microphones. Consistent gains over standard clustering and adaptation were obtained. Copyright © 2011 ISCA.

Use of contexts in language model interpolation and adaptation

Language models (LMs) are often constructed by building multiple individual component models that are combined using context independent interpolation weights. By tuning these weights, using either perplexity or discriminative approaches, it is possible to adapt LMs to a particular task. This paper investigates the use of context dependent weighting in both interpolation and test-time adaptation of language models. Depending on the previous word contexts, a discrete history weighting function is used to adjust the contribution from each component model. As this dramatically increases the number of parameters to estimate, robust weight estimation schemes are required. Several approaches are described in this paper. The first approach is based on MAP estimation where interpolation weights of lower order contexts are used as smoothing priors. The second approach uses training data to ensure robust estimation of LM interpolation weights. This can also serve as a smoothing prior for MAP adaptation. A normalized perplexity metric is proposed to handle the bias of the standard perplexity criterion to corpus size. A range of schemes to combine weight information obtained from training data and test data hypotheses are also proposed to improve robustness during context dependent LM adaptation. In addition, a minimum Bayes' risk (MBR) based discriminative training scheme is also proposed. An efficient weighted finite state transducer (WFST) decoding algorithm for context dependent interpolation is also presented. The proposed technique was evaluated using a state-of-the-art Mandarin Chinese broadcast speech transcription task. Character error rate (CER) reductions up to 7.3 relative were obtained as well as consistent perplexity improvements. © 2012 Elsevier Ltd. All rights reserved.

Language model cross adaptation for LVCSR system combination

State-of-the-art large vocabulary continuous speech recognition (LVCSR) systems often combine outputs from multiple sub-systems that may even be developed at different sites. Cross system adaptation, in which model adaptation is performed using the outputs from another sub-system, can be used as an alternative to hypothesis level combination schemes such as ROVER. Normally cross adaptation is only performed on the acoustic models. However, there are many other levels in LVCSR systems' modelling hierarchy where complimentary features may be exploited, for example, the sub-word and the word level, to further improve cross adaptation based system combination. It is thus interesting to also cross adapt language models (LMs) to capture these additional useful features. In this paper cross adaptation is applied to three forms of language models, a multi-level LM that models both syllable and word sequences, a word level neural network LM, and the linear combination of the two. Significant error rate reductions of 4.0-7.1% relative were obtained over ROVER and acoustic model only cross adaptation when combining a range of Chinese LVCSR sub-systems used in the 2010 and 2011 DARPA GALE evaluations. © 2012 Elsevier Ltd. All rights reserved.

Transducer and base compliance alter the in situ 6 dof force measured from muscle during an isometric contraction in a multi-joint limb.

Although musculoskeletal models are commonly used, validating the muscle actions predicted by such models is often difficult. In situ isometric measurements are a possible solution. The base of the skeleton is immobilized and the endpoint of the limb is rigidly attached to a 6-axis force transducer. Individual muscles are stimulated and the resulting forces and moments recorded. Such analyses generally assume idealized conditions. In this study we have developed an analysis taking into account the compliances due to imperfect fixation of the skeleton, imperfect attachment of the force transducer, and extra degrees of freedom (dof) in the joints that sometimes become necessary in fixed end contractions. We use simulations of the rat hindlimb to illustrate the consequences of such compliances. We show that when the limb is overconstrained, i.e., when there are fewer dof within the limb than are restrained by the skeletal fixation, the compliances of the skeletal fixation and of the transducer attachment can significantly affect measured forces and moments. When the limb dofs and restrained dofs are matched, however, the measured forces and moments are independent of these compliances. We also show that this framework can be used to model limb dofs, so that rather than simply omitting dofs in which a limb does not move (e.g., abduction at the knee), the limited motion of the limb in these dofs can be more realistically modeled as a very low compliance. Finally, we discuss the practical implications of these results to experimental measurements of muscle actions.

Is titin a 'winding filament'? A new twist on muscle contraction.

Recent studies have demonstrated a role for the elastic protein titin in active muscle, but the mechanisms by which titin plays this role remain to be elucidated. In active muscle, Ca(2+)-binding has been shown to increase titin stiffness, but the observed increase is too small to explain the increased stiffness of parallel elastic elements upon muscle activation. We propose a 'winding filament' mechanism for titin's role in active muscle. First, we hypothesize that Ca(2+)-dependent binding of titin's N2A region to thin filaments increases titin stiffness by preventing low-force straightening of proximal immunoglobulin domains that occurs during passive stretch. This mechanism explains the difference in length dependence of force between skeletal myofibrils and cardiac myocytes. Second, we hypothesize that cross-bridges serve not only as motors that pull thin filaments towards the M-line, but also as rotors that wind titin on the thin filaments, storing elastic potential energy in PEVK during force development and active stretch. Energy stored during force development can be recovered during active shortening. The winding filament hypothesis accounts for force enhancement during stretch and force depression during shortening, and provides testable predictions that will encourage new directions for research on mechanisms of muscle contraction.

Tandem system adaptation using multiple linear feature transforms

Adaptation to speaker and environment changes is an essential part of current automatic speech recognition (ASR) systems. In recent years the use of multi-layer percpetrons (MLPs) has become increasingly common in ASR systems. A standard approach to handling speaker differences when using MLPs is to apply a global speaker-specific constrained MLLR (CMLLR) transform to the features prior to training or using the MLP. This paper considers the situation when there are both speaker and channel, communication link, differences in the data. A more powerful transform, front-end CMLLR (FE-CMLLR), is applied to the inputs to the MLP to represent the channel differences. Though global, these FE-CMLLR transforms vary from time-instance to time-instance. Experiments on a channel distorted dialect Arabic conversational speech recognition task indicates the usefulness of adapting MLP features using both CMLLR and FE-CMLLR transforms. © 2013 IEEE.

Changes in biomechanical properties of the coronary artery wall contribute to maintained contractile responses to endothelin-1 in atherosclerosis.

AIMS: Our aim was to determine whether alterations in biomechanical properties of human diseased compared to normal coronary artery contribute to changes in artery responsiveness to endothelin-1 in atherosclerosis. MAIN METHODS: Concentration-response curves were constructed to endothelin-1 in normal and diseased coronary artery. The passive mechanical properties of arteries were determined using tensile ring tests from which finite element models of passive mechanical properties of both groups were created. Finite element modelling of artery endothelin-1 responses was then performed. KEY FINDINGS: Maximum responses to endothelin-1 were significantly attenuated in diseased (27±3 mN, n=55) compared to normal (38±2 mN, n=68) artery, although this remained over 70% of control. There was no difference in potency (pD2 control=8.03±0.06; pD2 diseased=7.98±0.06). Finite element modelling of tensile ring tests resulted in hyperelastic shear modulus μ=2004±410 Pa and hardening exponent α=22.8±2.2 for normal wall and μ=2464±1075 Pa and α=38.3±6.7 for plaque tissue and distensibility of diseased vessels was decreased. Finite element modelling of active properties of both groups resulted in higher muscle contractile strain (represented by thermal reactivity) of the atherosclerotic artery model than the normal artery model. The models suggest that a change in muscle response to endothelin-1 occurs in atherosclerotic artery to increase its distensibility towards that seen in normal artery. SIGNIFICANCE: Our data suggest that an adaptation occurs in medial smooth muscle of atherosclerotic coronary artery to maintain distensibility of the vessel wall in the presence of endothelin-1. This may contribute to the vasospastic effect of locally increased endothelin-1 production that is reported in this condition.