HI,欢迎来到学术之家,发表咨询:400-888-7501  订阅咨询:400-888-7502  股权代码  102064
0

Nonlinear Prediction with Deep Recurrent Neural Networks for Non-Blind Audio Bandwidth Extension

作者:Lin; Jiang; Ruimin; Hu; Xiaochen; Wang...audiocodingaudiobandwidthextensioncontextcorrelationdeeprecurrentneuralnetwork

摘要:Non-blind audio bandwidth exten-sion is a standard technique within contem-porary audio codecs to efficiently code audiosignals at low bitrates. In existing methods, inmost cases high frequencies signal is usuallygenerated by a duplication of the correspond-ing low frequencies and some parameters ofhigh frequencies. However, the perceptionquality of coding will significantly degradeif the correlation between high frequenciesand low frequencies becomes weak. In thispaper, we quantitatively analyse the correla-tion via computing mutual information value.The analysis results show the correlation alsoexists in low frequency signal of the contextdependent frames besides the current frame.In order to improve the perception quality ofcoding, we propose a novel method of highfrequency coarse spectrum generation to im-prove the conventional replication method. Inthe proposed method, the coarse high frequen-cy spectrums are generated by a nonlinearmapping model using deep recurrent neuralnetwork. The experiments confirm that theproposed method shows better performancethan the reference methods.

注:因版权方要求,不能公开全文,如需全文,请咨询杂志社

中国通信

《中国通信》(CN:11-5439/TN)是一本有较高学术价值的大型月刊,自创刊以来,选题新奇而不失报道广度,服务大众而不失理论高度。颇受业界和广大读者的关注和好评。

杂志详情