This paper presents an evaluation of parallel voice conversion (VC) with neural network (NN)-based statistical models for spectral mapping and waveform generation. The NN-based architectures for spectral mapping include deep NN (DNN), deep mixture density network (DMDN), and recurrent NN (RNN) models. WaveNet (WN) vocoder is employed as a high-quality NN-based waveform generation. In VC, though, owing to the oversmoothed characteristics of estimated speech parameters, quality degradation still occurs. To address this problem, we utilize post-conversion for the converted features based on direct waveform modifferential and global variance postfilter. To preserve the consistency with the post-conversion, we further propose a spectrum differential loss for the spectral modeling. The experimental results demonstrate that: (1) the RNN-based spectral modeling achieves higher accuracy with a faster convergence rate and better generalization compared to the DNN-/DMDN-based models; (2) the RNN-based spectral modeling is also capable of producing less oversmoothed spectral trajectory; (3) the use of proposed spectrum differential loss improves the performance in the same-gender conversions; and (4) the proposed post-conversion on converted features for the WN vocoder in VC yields the best performance in both naturalness and speaker similarity compared to the conventional use of WN vocoder.

The high beam current and subangstrom resolution of aberration-corrected scanning transmission electron microscopes has enabled electron energy loss spectroscopy (EELS) mapping with atomic resolution. These spectral maps are often dose limited and spatially oversampled, leading to low counts/channel and are thus highly sensitive to errors in background estimation. However, by taking advantage of redundancy in the dataset map, one can improve background estimation and increase chemical sensitivity. We consider two such approaches—linear combination of power laws and local background averaging—that reduce background error and improve signal extraction. Principal component analysis (PCA) can also be used to analyze spectrum images, but the poor peak-to-background ratio in EELS can lead to serious artifacts if raw EELS data are PCA filtered. We identify common artifacts and discuss alternative approaches. These algorithms are implemented within the Cornell Spectrum Imager, an open source software package for spectroscopic analysis.

For a connected open subset Ω of the plane and n a positive integer, let be the space introduced by Cowen and Douglas in their paper, “Complex geometry and operator theory”. Our main concern is the case n = 1, in which case we show the existence of a functional calculus for von Neumann operators in for which a spectral mapping theorem holds. In particular we prove that if the spectrum of , is a spectral set for T, and if , then σ(f(T)) = f(Ω)- for every bounded analytic function f on the interior of L, where L is compact, σ(T) ⊂ L, the interior of L is simply connected and L is minimal with respect to these properties. This functional calculus turns out to be nice in the sense that the general study of von Neumann operators in is reduced to the special situation where Ω is an open connected subset of the unit disc with .