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.

Assessment Of Dynamic Properties Of A Crowd Model For Human-Structure Interaction Modeling

As lightweight and slender structural elements are more frequently used in the design, large scale structures become more flexible and susceptible to excessive vibrations. To ensure the functionality of the structure, dynamic properties of the occupied structure need to be estimated during the design phase. Traditional analysis method models occupants simply as an additional mass; however, research has shown that human occupants could be better modeled as an additional degree-of- freedom. In the United Kingdom, active and passive crowd models are proposed by the Joint Working Group as a result of a series of analytical and experimental research. It is expected that the crowd models would yield a more accurate estimation to the dynamic response of the occupied structure. However, experimental testing recently conducted through a graduate student project at Bucknell University indicated that the proposed passive crowd model might be inaccurate in representing the impact on the structure from the occupants. The objective of this study is to provide an assessment of the validity of the crowd models proposed by JWG through comparing the dynamic properties obtained from experimental testing data and analytical modeling results. The experimental data used in this study was collected by Firman in 2010. The analytical results were obtained by performing a time-history analysis on a finite element model of the occupied structure. The crowd models were created based on the recommendations from the JWG combined with the physical properties of the occupants during the experimental study. During this study, SAP2000 was used to create the finite element models and to implement the analysis; Matlab and ME¿scope were used to obtain the dynamic properties of the structure through processing the time-history analysis results from SAP2000. The result of this study indicates that the active crowd model could quite accurately represent the impact on the structure from occupants standing with bent knees while the passive crowd model could not properly simulate the dynamic response of the structure when occupants were standing straight or sitting on the structure. Future work related to this study involves improving the passive crowd model and evaluating the crowd models with full-scale structure models and operating data.

Performance Modeling Of Reconfigurable Distributed Systems Based On The Opensparc Fpga Board And The Sirc Communication Framewor

This thesis explores system performance for reconfigurable distributed systems and provides an analytical model for determining throughput of theoretical systems based on the OpenSPARC FPGA Board and the SIRC Communication Framework. This model was developed by studying a small set of variables that together determine a system¿s throughput. The importance of this model is in assisting system designers to make decisions as to whether or not to commit to designing a reconfigurable distributed system based on the estimated performance and hardware costs. Because custom hardware design and distributed system design are both time consuming and costly, it is important for designers to make decisions regarding system feasibility early in the development cycle. Based on experimental data the model presented in this paper shows a close fit with less than 10% experimental error on average. The model is limited to a certain range of problems, but it can still be used given those limitations and also provides a foundation for further development of modeling reconfigurable distributed systems.