Modeling Dynamics In Connectionist Speech Recognition The Time Index Model

Modeling Dynamics in Connectionist Speech Recognition : the Time Index Model

Abstract: "Here, we introduce an alternative to the Hidden Markov Model (HMM) as the underlying representation of speech production. HMMs suffer from well known limitations, such as the unrealistic assumption that the observations generated in a given state are independent and identically distributed (i.i.d). We propose a time index model that explicitly conditions the emission probability of a state on the time index, i.e., on the number of 'visits' in the current state of the Markov chain in a sequence. Thus, the proposed model does not require an i.i.d. assumption. The connectionist framework enables us to represent the dependence on the time index as a non-parametric distribution and to share parameters between different speech unit models. Furthermore, we discuss an extension to the basic time index model by incorporating information about the duration of the phone segments. Our initial results show that given the position of the boundaries between basic speech units, e.g., phones, we can improve our current connectionist system performance significantly by using this model. However, we still do not know whether these boundaries can be estimated reliably, nor do we know how much benefit we can obtain from this method given less accurate boundary information. Currently we are experimenting with two possible approaches: trying to learn smooth probability densities for the boundaries, and getting a set of reasonable segmentations from an N-Best search. In both cases we will need to consider the effect of incorrect boundaries, since they will undoubtedly occur."

Advances in Neural Information Processing Systems 8

The past decade has seen greatly increased interaction between theoretical work in neuroscience, cognitive science and information processing, and experimental work requiring sophisticated computational modeling. The 152 contributions in NIPS 8 focus on a wide variety of algorithms and architectures for both supervised and unsupervised learning. They are divided into nine parts: Cognitive Science, Neuroscience, Theory, Algorithms and Architectures, Implementations, Speech and Signal Processing, Vision, Applications, and Control. Chapters describe how neuroscientists and cognitive scientists use computational models of neural systems to test hypotheses and generate predictions to guide their work. This work includes models of how networks in the owl brainstem could be trained for complex localization function, how cellular activity may underlie rat navigation, how cholinergic modulation may regulate cortical reorganization, and how damage to parietal cortex may result in neglect. Additional work concerns development of theoretical techniques important for understanding the dynamics of neural systems, including formation of cortical maps, analysis of recurrent networks, and analysis of self- supervised learning. Chapters also describe how engineers and computer scientists have approached problems of pattern recognition or speech recognition using computational architectures inspired by the interaction of populations of neurons within the brain. Examples are new neural network models that have been applied to classical problems, including handwritten character recognition and object recognition, and exciting new work that focuses on building electronic hardware modeled after neural systems. A Bradford Book

REMAP