Adrian Corduneanu

Correlation modeling of MLLR transform biases for rapid HMM adaptation to new speakers

This paper concerns rapid adaptation of hidden Markov model (HMM) based speech recognizers to a new speaker, when only few speech samples (one minute or less) are available from the new speaker. A widely used family of adaptation algorithms defines adaptation as a linearly constrained reestimation of the HMM Gaussians. With few speech data, tight constraints must be introduced, by reducing the number of linear transforms and by specifying certain transform structures (e.g. block diagonal). We hypothesize that under these adaptation conditions, the residual errors of the adapted Gaussian parameters can be represented and corrected by dependency models, as estimated from a training corpus. Thus, after introducing a particular class of linear transforms, we develop correlation models of the transform parameters. In rapid adaptation experiments on the Switchboard corpus, the proposed algorithm performs better than the transform-constrained adaptation and the adaptation by correlation modeling of the HMM parameters, respectively.

Learning spatially-variable filters for super-resolution of text

Images magnified by standard methods display a degradation of detail that is particularly noticeable in the blurry edges of text. Current super-resolution algorithms address the lack of sharpness by filling in the image with probable details. These algorithms break the outlines of text. Our novel algorithm for super-resolution of text magnifies images in real-time by interpolation with a variable linear filter. The coefficients of the filter are determined nonlinearly from the neighborhood to which it is applied. We train the mapping that defines the coefficients to specifically enhance edges of text, producing a conservative algorithm that infers the detail of magnified text. Possible applications include resizing web page layouts or other interfaces, and enhancing low resolution camera captures of text. In general, learning spatially-variable filters is applicable to other image filtering tasks.

A Pylonic Decision-Tree Language Model- with Optimal Question Selection

This paper discusses a decision-tree approach to the problem of assigning probabilities to words following a given text. In contrast with previous decision-tree language model attempts, an algorithm for selecting nearly optimal questions is considered. The model is to be tested on a standard task, The Wall Street Journal, allowing a fair comparison with the well-known trigram model.