We present a large vocabulary, continuous speech recognition system based on Linked Predictive Neural Networks (LPNN's). The system is a straightforward extension, from isolated to continuous speech, of the LPNN system presented last year. The system uses neural networks as predictors of speech frames, yielding distortion measures which can be used by the One Stage DTW algorithm to perform continuous speech recognition. The system currently achieves 95%, 58%, and 39% word accuracy on tasks with perplexity 7, 11 1, and 402 respectively, outperforming several simple HMMs that we tested. We also found that the accuracy and speed of the LPNN can be slightly improved by the judicious use of hidden control inputs. We conclude by discussing the strengths and weaknesses of the predictive approach.

No abstract

Electron holography to image magnetic domains (invited)We describe a laser magneto-optic microscope (LAMOM) system, used to observe magnetic domain dynamics in Permalloy thin-film devices. Optimization of Kerr magneto-optic contrast is achieved by using laser illumination in a polarized light microscope in combination with video image processing. High magnetic contrast, a diffraction-limited (0.25 pm) resolution, and motion picture recording capability at up to 30 frames/s are demonstrated by recent results.

The results from applying an improved algorithm in the task of automatic segmentation of spontaneoustelephone quality speech are presented, and compared to the results from those resulting from superimposing white noise. Three segmentation algorithms are compared which are all based on variants of the Spectral Variation Function. Experimental results are obtained on the OGI multi-language telephone speech corpus (OGI TS). We show that the use of the auditory forward and backward masking effects prior to the SVF computation increases the robustness of the algorithm to white noise. When the average signal-to-noise ratio (SNR) is decreased to 10dB the peak ratio (defined as the ratio of the number of peaks measured at the target over the original SNRs) is increased by 16%, 12%, and 11% for the MFC (Mel-Frequency Cepstra), RASTA (RelAtive SpecTrAl processing), and the FBDYN (Forward-Backward auditory masking DYNamic cepstra) SVF segmentation algorithms, respectively.

AbsWnct -Wide-field magneto-optic Kerr observation techniques are described for the imaging of magnetic recording heads and other small magnetic devices. The imaging prohlems encountered for these samples are quite different from those for continuous films and lead to a dimerent set of experimental approaches.Ohservation techniques that preserve the magnetic state of the sample allow for the implementation of new image acquisition algorithms based on multiple additions and subtractions that progressively build up a weak Kerr contrast even under unfavorable experimental conditions. Examples of results are presented.