In this paper, we propose a method of enhancing whisper, using whisper without any pretreatment combined with Wavenet. Our method is end-to-end, that is, inputing noised whisper to get clean whisper. The input to our ...
In this paper, we propose a method of enhancing whisper, using whisper without any pretreatment combined with Wavenet. Our method is end-to-end, that is, inputing noised whisper to get clean whisper. The input to our method is the original whisper without any processing, reducing the loss of features caused by other operations. We use speech denoising Wavenet to enhance whisper. Wavenet can not only enhance whisper well, but also tackle the issue of intelligibility. Specifically, use symmetric dilated convolution to obtain noisy speech context, help the model to enhance the speech for better denoising effect. Experimental results show that the enchanced whisper gains better performance both in the aspect of speech quality and intelligibility.
In this paper, we are interested in the conversion of whispered to normal speech. The baseline method uses standard bidirectional LSTM (BLSTM) RNN to predict both the spectral features and excitation parameters of the...
In this paper, we are interested in the conversion of whispered to normal speech. The baseline method uses standard bidirectional LSTM (BLSTM) RNN to predict both the spectral features and excitation parameters of the normal speech from whispered speech. Also, it employs STRAIGHT speech synthesizer. The BLSTM based whispered speech to normal speech conversion system is among the best systems in term of the naturalness of generated speech. However, in many cases, the model complexity and inference cost of BLSTM prevents its usage. As opposed to using standard BLSTM with sharing values, we propose a meta-network to generate non-shared weights for LSTM memory block in BLSTM (denote as meta-BLSTM). Besides, we use a low-rank approximation to generate the parameter matrix, which can reduce the model complexity. To our knowledge, this is the first study that uses meta-network to train a whispered to normal speech conversion system. To evaluate the performance of the proposed system, we performed experiments in the TIMIT dataset. Experimental results show that the proposed method achieves state-of-the-art performance.
Abstract-Brain-computer interface (BCI) which transforms signals from the brain into control signals can help people with disabilities communicate with others. In this paper, posteriori probability support vector mach...
详细信息
Abstract-Brain-computer interface (BCI) which transforms signals from the brain into control signals can help people with disabilities communicate with others. In this paper, posteriori probability support vector machine (PPSVM) for patterns recognition was developed. For the classification of the left or right hand motor imagery, this method was used to expend the training set by adding samples with great probability output. For the dataset from 2003 BCI Competition, AR model was adopted to extract feature vectors and SVM with posteriori probabilistic output was used to classify the dataset. The results proved that, by adding samples with big probability, the performance of BCI was improved and higher accuracy was achieved.
In this paper, a deep learning optimization method combining U-net model and cycle generative adversarial network (Cycle-GAN) is proposed to efficiently solve the electromagnetic inverse scattering (EMIS) problems. Fi...
详细信息
A multi-factor model for the grain yield prediction based on the previous data and relevant impacting factors is reported. In this model, the weight coefficient of each individual factor that affected the grain yield ...
详细信息
A multi-factor model for the grain yield prediction based on the previous data and relevant impacting factors is reported. In this model, the weight coefficient of each individual factor that affected the grain yield historically is analyzed by the variation coefficient method and the conversion degree function in the attribute theory. The effects of various factors on the predicted grain yield are then used as standard vectors and subjected to a similarity-based search in the matrix of historical values. The predicted total grain yield is then determined by multiplying the sown area by the unit grain yield which is obtained by the highest similarity between the historical data and those predicted. Compared to the results obtained by the BP nerve network method, this method is simpler, more flexible, less time-consuming, and more accurate.
We analyzed the seismic waveforms from the December 26, 2004 Sumatra-Andaman earthquake recorded at broadband seismic stations in western Europe. Previous studies involving of the beam-forming technique and high frequ...
详细信息
We analyzed the seismic waveforms from the December 26, 2004 Sumatra-Andaman earthquake recorded at broadband seismic stations in western Europe. Previous studies involving of the beam-forming technique and high frequency analysis suggest that the earthquake ruptured with a duration of around 500 s. This very long duration makes P wave overlap with later arrivals such as PP wave, which follows P in about 200 s. Since P waves are crucial for modeling earthquake processes, we propose an iterative method to separate P and PP waveforms. The separated P waveform confirms a second large energy release around 300 s after the initial rupture. The iterative signal separation technique is particularly useful for mixed signals that are not independent and the number of recording stations far exceeds number of mixed signal sources.
The paper presents a novel circular polarization(CP) antenna loading with a parasitic ring metal strip, which is designed for global positioning system (GPS) L1 band applications. The antenna consists of a defected gr...
详细信息
As an important research branch,memristor has attracted a range of scholars to study the property of memristive chaotic ***,time⁃delayed systems are considered a significant and newly⁃developing field in modern *** co...
详细信息
As an important research branch,memristor has attracted a range of scholars to study the property of memristive chaotic ***,time⁃delayed systems are considered a significant and newly⁃developing field in modern *** combining memristor and time⁃delay,a delayed memristive differential system with fractional order is proposed in this paper,which can generate hidden ***,we discussed the dynamics of the proposed system where the parameter was set as the bifurcation parameter,and showed that with the increase of the parameter,the system generated rich chaotic phenomena such as bifurcation,chaos,and *** we derived adequate and appropriate stability criteria to guarantee the system to achieve ***,examples were provided to analyze and confirm the influence of parameter a,fractional order q,and time delayτon chaos *** simulation results confirm that the chaotic synchronization is affected by a,q andτ.
The theory of granule computing based on the quotient space is one of the three main granule computing theories. The emphasis is on the structure of the quotient space theory in this paper. Comparing with Rough Set th...
详细信息
作者:
Zhang, YanpingWang, YuehuaZhao, ShuAnhui University
Computer Science and Technology Institute Key Laboratory of Intelligent Computing and Signal Processing of Ministry of Education Hefei Anhui Province 230039 China
In this paper, we propose a novel distributed machine learning method: Parallel Covering Algorithm, which is inspired by the module feature of CA (Covering Algorithm). Classic method of CA is presented, and we analyze...
详细信息
暂无评论