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.

Changes in hemodynamics and tissue oxygenation saturation in the brain and skeletal muscle induced by speech therapy - a near-infrared spectroscopy study

Arts speech therapy (AST) is a therapeutic method within complementary medicine and has been practiced for decades for various medical conditions. It comprises listening and the recitation of different forms of speech exercises under the guidance of a licensed speech therapist. The aim of our study was to noninvasively investigate whether different types of recitation influence hemodynamics and oxygenation in the brain and skeletal leg muscle using near-infrared spectroscopy (NIRS). Seventeen healthy volunteers (eight men and nine women, mean age ± standard deviation 35.6 ± 12.7 years) were enrolled in the study. Each subject was measured three times on different days with the different types of recitation: hexameter, alliteration, and prose verse. Before, during, and after recitation, relative concentration changes of oxyhemoglobin (Δ[O2Hb]), deoxyhemoglobin (Δ[HHb]), total hemoglobin (Δ[tHb]), and tissue oxygenation saturation (StO2) were measured in the brain and skeletal leg muscle using a NIRS device. The study was performed with a randomized crossover design. Significant concentration changes were found during recitation of all verses, with mainly a decrease in Δ[O2Hb] and ΔStO2 in the brain, and an increase in Δ[O2Hb] and Δ[tHb] in the leg muscle during recitation. After the recitations, significant changes were mainly increases of Δ[HHb] and Δ[tHb] in the calf muscle. The Mayer wave spectral power (MWP) was also significantly affected, i.e., mainly the MWP of the Δ[O2Hb] and Δ[tHb] increased in the brain during recitation of hexameter and prose verse. The changes in MWP were also significantly different between hexameter and alliteration, and hexameter and prose. Possible physiological explanations for these changes are discussed. A probable reason is a different effect of recitations on the sympathetic nervous system. In conclusion, these changes show that AST has relevant effects on the hemodynamics and oxygenation of the brain and muscle.