Szu-Wei Fu

  • Citations Per Year
Learn More
This paper proposes a signal-to-noise-ratio (SNR) aware convolutional neural network (CNN) model for speech enhancement (SE). Because the CNN model can deal with local temporal-spectral structures of speech signals, it can effectively disentangle the speech and noise signals given the noisy speech signals. In order to enhance the generalization capability(More)
In vitro modulation of the differentiation status of mesenchymal stem cells (MSCs) is important for their application to regenerative medicine. We suggested that the morphology and differentiation states of MSCs could be modulated by controlling the cell affinity of a substrate. The objective of this study was to investigate the effects of surface(More)
This study proposes a fully convolutional network (FCN) model for raw waveform-based speech enhancement. The proposed system performs speech enhancement in an end-to-end (i.e., waveform-in and waveform-out) manner, which differs from most existing denoising methods that process the magnitude spectrum (e.g., log power spectrum (LPS)) only. Because the fully(More)
Transplantation of cell sheets including an intact extracellular matrix is one tissue-engineering strategy for tissue regeneration. Temperature-responsive substrates based on poly(N-isopropylacrylamide) (PNIPAAm) have been used to harvest intact cell sheets by temperature change. In this work, we immobilized PNIPAAm on plastic substrates by a UV-activated(More)
In image deblurring, it is important to reconstruct images with small error, high perception quality, and less computational time. In this paper, a blurred image reconstruction algorithm, which is a combination of the Richardson-Lucy (RL) deconvolution approach and a pyramid structure, is proposed. The RL approach has good performance in image(More)
Speech enhancement model is used to map a noisy speech to a clean speech. In the training stage, an objective function is often adopted to optimize the model parameters. However, in most studies, there is an inconsistency between the model optimization criterion and the evaluation criterion on the enhanced speech. For example, in measuring speech(More)
Conventionally, the maximum likelihood (ML) criterion is applied to train a deep belief network (DBN). We present a maximum entropy (ME) learning algorithm for DBNs, designed specifically to handle limited training data. Maximizing only the entropy of parameters in the DBN allows more effective generalization capability, less bias towards data(More)
Feature points, such as SIFT, BRISK, ORB, and FREAK, are effective for template matching, pattern recognition, and object alignment. However, since an image usually has 200-4000 feature points and the size of each descriptor is 512 or 256, an efficient way for encoding the descriptors and locations of feature points is required. In this paper, we propose an(More)
The stroke is a very important feature for a character and is helpful for word recognition and handwriting identification. Although thinning algorithms can be applied for stroke extraction, they always suffer from the problems of bifurcation and disconnection. Moreover, since the end points of strokes cannot be preserved by thinning, the stroke length(More)
Objective: This paper focuses on machine learning based voice conversion (VC) techniques for improving the speech intelligibility of surgical patients who have had parts of their articulators removed. Because of the removal of parts of the articulator, a patient's speech may be distorted and difficult to understand. To overcome this problem, VC methods can(More)