It is well known that automatic speech recognition(ASR) is a resource consuming task. It takes sufficient amount of data to train a state-of-the-art deep neural network acoustic model. As for some low-resource languag...It is well known that automatic speech recognition(ASR) is a resource consuming task. It takes sufficient amount of data to train a state-of-the-art deep neural network acoustic model. As for some low-resource languages where scripted speech is difficult to obtain, data sparsity is the main problem that limits the performance of speech recognition system. In this paper, several knowledge transfer methods are investigated to overcome the data sparsity problem with the help of high-resource languages.The first one is a pre-training and fine-tuning(PT/FT) method, in which the parameters of hidden layers are initialized with a welltrained neural network. Secondly, the progressive neural networks(Prognets) are investigated. With the help of lateral connections in the network architecture, Prognets are immune to forgetting effect and superior in knowledge transferring. Finally,bottleneck features(BNF) are extracted using cross-lingual deep neural networks and serves as an enhanced feature to improve the performance of ASR system. Experiments are conducted in a low-resource Vietnamese dataset. The results show that all three methods yield significant gains over the baseline system, and the Prognets acoustic model performs the best. Further improvements can be obtained by combining the Prognets model and bottleneck features.展开更多
The cDNA insert of the plasmid p14-6[1] is found to be the 3’-untranslatcd region (3’-UTR) of the transcription factor for human interleukin-6, NF-IL6. This 3’ -DTK is actively transcribed in the revertant cell lin...The cDNA insert of the plasmid p14-6[1] is found to be the 3’-untranslatcd region (3’-UTR) of the transcription factor for human interleukin-6, NF-IL6. This 3’ -DTK is actively transcribed in the revertant cell line RR, which contains the p14-6 plasmid integrated into its genomic DNA. Simultaneously a protein specifically bound to this 3’-UTR is expressed in significantly larger amounts. Its overexpression is apparently related to the reversion of the malignant cellular phenotype. The properties of this protein, named BNF, and possible reasons for its overexpression are discussed, and hypothesis on the mechanism of reversion of the RR cells is proposed.展开更多
Meta-modelling plays an important role in model driven software development. In this paper, a graphic exten- sion of BNF （GEBNF） is proposed to define the abstract syn- tax of graphic modelling languages. From a GEB...Meta-modelling plays an important role in model driven software development. In this paper, a graphic exten- sion of BNF （GEBNF） is proposed to define the abstract syn- tax of graphic modelling languages. From a GEBNF syntax definition, a formal predicate logic language can be induced so that meta-modelling can be performed formally by spec- ifying a predicate on the domain of syntactically valid mod- els. In this paper, we investigate the theoretical foundation of this meta-modelling approach. We formally define the se- mantics of GEBNF and its induced predicate logic languages, then apply Goguen and Burstall＇s institution theory to prove that they form a sound and valid formal specification lan- guage for meta-modelling.展开更多
基金partially supported by the National Natural Science Foundation of China(11590770-4,U1536117)the National Key Research and Development Program of China(2016YFB0801203,2016YFB0801200)+1 种基金the Key Science and Technology Project of the Xinjiang Uygur Autonomous Region(2016A03007-1)the Pre-research Project for Equipment of General Information System(JZX2017-0994/Y306)
文摘It is well known that automatic speech recognition(ASR) is a resource consuming task. It takes sufficient amount of data to train a state-of-the-art deep neural network acoustic model. As for some low-resource languages where scripted speech is difficult to obtain, data sparsity is the main problem that limits the performance of speech recognition system. In this paper, several knowledge transfer methods are investigated to overcome the data sparsity problem with the help of high-resource languages.The first one is a pre-training and fine-tuning(PT/FT) method, in which the parameters of hidden layers are initialized with a welltrained neural network. Secondly, the progressive neural networks(Prognets) are investigated. With the help of lateral connections in the network architecture, Prognets are immune to forgetting effect and superior in knowledge transferring. Finally,bottleneck features(BNF) are extracted using cross-lingual deep neural networks and serves as an enhanced feature to improve the performance of ASR system. Experiments are conducted in a low-resource Vietnamese dataset. The results show that all three methods yield significant gains over the baseline system, and the Prognets acoustic model performs the best. Further improvements can be obtained by combining the Prognets model and bottleneck features.
基金Project supported by the 863 Program,National Science and Technology Commission, China.
文摘The cDNA insert of the plasmid p14-6[1] is found to be the 3’-untranslatcd region (3’-UTR) of the transcription factor for human interleukin-6, NF-IL6. This 3’ -DTK is actively transcribed in the revertant cell line RR, which contains the p14-6 plasmid integrated into its genomic DNA. Simultaneously a protein specifically bound to this 3’-UTR is expressed in significantly larger amounts. Its overexpression is apparently related to the reversion of the malignant cellular phenotype. The properties of this protein, named BNF, and possible reasons for its overexpression are discussed, and hypothesis on the mechanism of reversion of the RR cells is proposed.
文摘Meta-modelling plays an important role in model driven software development. In this paper, a graphic exten- sion of BNF （GEBNF） is proposed to define the abstract syn- tax of graphic modelling languages. From a GEBNF syntax definition, a formal predicate logic language can be induced so that meta-modelling can be performed formally by spec- ifying a predicate on the domain of syntactically valid mod- els. In this paper, we investigate the theoretical foundation of this meta-modelling approach. We formally define the se- mantics of GEBNF and its induced predicate logic languages, then apply Goguen and Burstall＇s institution theory to prove that they form a sound and valid formal specification lan- guage for meta-modelling.
