“…For V/NV detection voiced and nonvoiced regions of speech are considered for learning the dictionaries. NMF based methods (such as in Teng and Jia (2013) and Franois et al (2013)) are batch algorithms which perform well, but relies on pre-estimation of dictionaries for both noise and clean speech signal for efficient sparse representation, which might not be a suitable choice in the practical scenario. The supervised method proposed in Saeedi et al (2013) employ SVM models trained in different background noises for speech/non-speech classification.…”
Section: Methodsmentioning
confidence: 99%
“…Recent works in Teng and Jia (2013) and You et al (2012) have proposed methods based on sparse coding for VAD rather than V/NV detection. These methods are discussed here mainly due to their similarity in using a sparse coding framework (but on speech signal directly).…”
Section: Background and Prior Workmentioning
confidence: 99%
“…However, both the methods require several examples of clean speech to learn the dictionary. Moreover, the method in Teng and Jia (2013) based on non-negative matrix factorization (NMF), employs a separate dictionary for noise examples, which is not suitable in practical scenarios, as generally the type of noise is not known a priori. Method in Teng and Jia (2013) uses the maximum and mean value, while method in You et al (2012) uses power spectrum energy of the estimated sparse representation as a VAD feature, which might not be robust in low SNR conditions, and in the presence of highly non-stationary noises having sparse or speech like characteristics.…”
Section: Background and Prior Workmentioning
confidence: 99%
“…Therefore, it is very difficult to detect unvoiced regions as compared to voiced regions in the presence of noise. This is also the reason that recently proposed sparse coding based VAD algorithms (such as in Teng and Jia (2013) and You et al (2012)) have poor performance in detecting unvoiced regions.…”
Section: Source Characteristics Using Sparse Vectormentioning
“…For V/NV detection voiced and nonvoiced regions of speech are considered for learning the dictionaries. NMF based methods (such as in Teng and Jia (2013) and Franois et al (2013)) are batch algorithms which perform well, but relies on pre-estimation of dictionaries for both noise and clean speech signal for efficient sparse representation, which might not be a suitable choice in the practical scenario. The supervised method proposed in Saeedi et al (2013) employ SVM models trained in different background noises for speech/non-speech classification.…”
Section: Methodsmentioning
confidence: 99%
“…Recent works in Teng and Jia (2013) and You et al (2012) have proposed methods based on sparse coding for VAD rather than V/NV detection. These methods are discussed here mainly due to their similarity in using a sparse coding framework (but on speech signal directly).…”
Section: Background and Prior Workmentioning
confidence: 99%
“…However, both the methods require several examples of clean speech to learn the dictionary. Moreover, the method in Teng and Jia (2013) based on non-negative matrix factorization (NMF), employs a separate dictionary for noise examples, which is not suitable in practical scenarios, as generally the type of noise is not known a priori. Method in Teng and Jia (2013) uses the maximum and mean value, while method in You et al (2012) uses power spectrum energy of the estimated sparse representation as a VAD feature, which might not be robust in low SNR conditions, and in the presence of highly non-stationary noises having sparse or speech like characteristics.…”
Section: Background and Prior Workmentioning
confidence: 99%
“…Therefore, it is very difficult to detect unvoiced regions as compared to voiced regions in the presence of noise. This is also the reason that recently proposed sparse coding based VAD algorithms (such as in Teng and Jia (2013) and You et al (2012)) have poor performance in detecting unvoiced regions.…”
Section: Source Characteristics Using Sparse Vectormentioning
“…In the aspect of voice recognition, sparse representation has achieved many successes [1][2][3][4][5][6][7][8][9]. G.S.V.S Sivaram et al [10] propose a novel feature extraction technique for speech recognition based on the principles of sparse coding.…”
In an ecological breeding environment, feeders usually don't know whether their chicken are in good condition. Fortunately, Chicken voices reveal a lot of messages and it is easy to access and process voice signals. Therefore, we propose a newtype chicken voice recognition method using sparse representation aiming at the chicken voice recognition problem in an ecological breeding environment. In future, the method can be used in chickens for automated disease detection during the period of the bird flu. First, we use a multi-band spectral subtraction method for de-noise processing. Second, we reconstruct voice signals via sparse representation using the orthogonal matching pursuit algorithm. Third, we extract Melfrequency cepstral coefficients (MFCC), linear predictive coding (LPC) and power-normalized cepstral coefficients (PNCC) from the chicken voices. Finally, we use support vector machine (SVM) to classify chicken voices under different environments. Extensive experimental results show that the features using sparse representation can respectively get better recognition effects.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.