Corpus-Based Speech Enhancement With Uncertainty Modeling and Cepstral Smoothing

We present a new approach for corpus-based speech enhancement that significantly improves over a method published by Xiao and Nickel in 2010. Corpus-based enhancement systems do not merely filter an incoming noisy signal, but resynthesize its speech content via an inventory of pre-recorded clean signals. The goal of the procedure is to perceptually improve the sound of speech signals in background noise. The proposed new method modifies Xiao's method in four significant ways. Firstly, it employs a Gaussian mixture model (GMM) instead of a vector quantizer in the phoneme recognition front-end. Secondly, the state decoding of the recognition stage is supported with an uncertainty modeling technique. With the GMM and the uncertainty modeling it is possible to eliminate the need for noise dependent system training. Thirdly, the post-processing of the original method via sinusoidal modeling is replaced with a powerful cepstral smoothing operation. And lastly, due to the improvements of these modifications, it is possible to extend the operational bandwidth of the procedure from 4 kHz to 8 kHz. The performance of the proposed method was evaluated across different noise types and different signal-to-noise ratios. The new method was able to significantly outperform traditional methods, including the one by Xiao and Nickel, in terms of PESQ scores and other objective quality measures. Results of subjective CMOS tests over a smaller set of test samples support our claims.

Vibrational Relaxation of 13CO2(v2) by Atomic Oxygen

Carbon dioxide (CO2) has been of recent interest due to the issue of greenhouse cooling in the upper atmosphere by species such as CO2 and NO. In the Earth’s upper atmosphere, between altitudes of 75 and 110 km, a collisional energy exchange occurs between CO2 and atomic oxygen, which promotes a population of ground state CO2 to the bend excited state. The relaxation of CO2 following this excitation is characterized by spontaneous emission of 15-μm. Most of this energy is emitted away from Earth. Due to the low density in the upper atmosphere, most of this energy is not reabsorbed and thus escapes into space, leading to a local cooling effect in the upper atmosphere. To determine the efficiency of the CO2- O atom collisional energy exchange, transient diode laser absorption spectroscopy was used to monitor the population of the first vibrationally excited state, 13CO2(0110) or ν2, as a function of time. The rate coefficient, kO(ν2), for the vibrational relaxation 13CO2 (ν2)-O was determined by fitting laboratory measurements using a home-written linear least squares algorithm. The rate coefficient, kO(ν2), of the vibrational relaxation of 13CO2(ν2), by atomic oxygen at room temperature was determined to be (1.6 ± 0.3 x 10-12 cm3 s-1), which is within the uncertainty of the rate coefficient previously found in this group for 12CO2(ν2) relaxation. The cold temperature kO(ν2) values were determined to be: (2.1 ± 0.8) x 10-12 cm3 s-1 at Tfinal = 274 K, (1.8 ± 0.3) x 10-12 cm3 s-1 at Tfinal = 239 K, (2 ± 1) x 10-12 cm3 s-1 at Tfinal = 208 K, and (1.7 ± 0.3) x 10-12 cm3 s-1 at Tfinal = 186 K. These data did not show a definitive negative temperature dependence comparable to that found for 12CO2 previously.

Application of Discrete Fourier Inter-Coefficient Difference for Assessing Genetic Sequence Similarity

Digital signal processing (DSP) techniques for biological sequence analysis continue to grow in popularity due to the inherent digital nature of these sequences. DSP methods have demonstrated early success for detection of coding regions in a gene. Recently, these methods are being used to establish DNA gene similarity. We present the inter-coefficient difference (ICD) transformation, a novel extension of the discrete Fourier transformation, which can be applied to any DNA sequence. The ICD method is a mathematical, alignment-free DNA comparison method that generates a genetic signature for any DNA sequence that is used to generate relative measures of similarity among DNA sequences. We demonstrate our method on a set of insulin genes obtained from an evolutionarily wide range of species, and on a set of avian influenza viral sequences, which represents a set of highly similar sequences. We compare phylogenetic trees generated using our technique against trees generated using traditional alignment techniques for similarity and demonstrate that the ICD method produces a highly accurate tree without requiring an alignment prior to establishing sequence similarity.