Indirect heating and intensified digestion technology can be applied to reduce greatly the energy consumption in Bayer process of diasporic bauxite. A great advantage of two-stream process is to avoid or reduce effici...Indirect heating and intensified digestion technology can be applied to reduce greatly the energy consumption in Bayer process of diasporic bauxite. A great advantage of two-stream process is to avoid or reduce efficiently serious scaling problem of bauxite slurry on indirect heating surface, which certainly happens in the single stream process and brings about great troubles to the indirect heating. As a result of a great number of experiments and the theoretical analysis, a new lime adding technology for the two-stream digestion process is developed in this paper that lime is added into spent liquor stream instead of bauxite slurry, which is more suitable to the two-stream process of diasporic bauxite. The influences of the new lime addition technology on preheating and digestion process were discussed. It was deduced that the new technology can be used efficiently in the two-stream process of non-diasporic bauxite.展开更多
汉语是一种有调语言,因此在汉语语音识别中,调型信息起着非常关键的作用。在现有的隐马尔可夫模型(Hidden Markov Model)框架下,如何有效地利用调型信息是有待研究的问题。现有的汉语语音识别系统中主要采用两种方式来使用调型信息:一...汉语是一种有调语言,因此在汉语语音识别中,调型信息起着非常关键的作用。在现有的隐马尔可夫模型(Hidden Markov Model)框架下,如何有效地利用调型信息是有待研究的问题。现有的汉语语音识别系统中主要采用两种方式来使用调型信息:一种是基于Embedded Tone Model,即将调型特征向量与声学特征向量组成一个流去训练模型;一种是Explicit Tone Model,即将调型信息单独建模,再利用此模型优化原有的解码网络。该文将两种方法统一起来,首先利用Embedded Tone Model采用双流而非单流建模得到Nbest备选,再利用Explicit ToneModel对调进行左相关建模并对Nbest得分重新修正以得到识别结果,从而获得性能提升。与传统的无调模型相比,该文方法的识别率的平均绝对提升超过了3.0%,在第三测试集上的绝对提升达到了5.36%。展开更多
文摘Indirect heating and intensified digestion technology can be applied to reduce greatly the energy consumption in Bayer process of diasporic bauxite. A great advantage of two-stream process is to avoid or reduce efficiently serious scaling problem of bauxite slurry on indirect heating surface, which certainly happens in the single stream process and brings about great troubles to the indirect heating. As a result of a great number of experiments and the theoretical analysis, a new lime adding technology for the two-stream digestion process is developed in this paper that lime is added into spent liquor stream instead of bauxite slurry, which is more suitable to the two-stream process of diasporic bauxite. The influences of the new lime addition technology on preheating and digestion process were discussed. It was deduced that the new technology can be used efficiently in the two-stream process of non-diasporic bauxite.
文摘汉语是一种有调语言,因此在汉语语音识别中,调型信息起着非常关键的作用。在现有的隐马尔可夫模型(Hidden Markov Model)框架下,如何有效地利用调型信息是有待研究的问题。现有的汉语语音识别系统中主要采用两种方式来使用调型信息:一种是基于Embedded Tone Model,即将调型特征向量与声学特征向量组成一个流去训练模型;一种是Explicit Tone Model,即将调型信息单独建模,再利用此模型优化原有的解码网络。该文将两种方法统一起来,首先利用Embedded Tone Model采用双流而非单流建模得到Nbest备选,再利用Explicit ToneModel对调进行左相关建模并对Nbest得分重新修正以得到识别结果,从而获得性能提升。与传统的无调模型相比,该文方法的识别率的平均绝对提升超过了3.0%,在第三测试集上的绝对提升达到了5.36%。
