摘要
提出一种音素相关特征,并将语言学中语支的思想引入语种识别。结合音素相关特征和因子分析方法,提出音素层语支变化量(PLBV)方法。通过对音素后验概率进行降维和均值方差规整,得到音素相关特征。使用因子分析技术将各语支变化量空间的低维变化量因子进行拼接得到音素层语支变化量因子,然后在语支内部和语支间分别对语支变化量因子进行支持向量机(SVM)建模。实验基于俄语音子识别器,在美国国家标准技术署(NIST)2011年语种识别评测(LRE)30s数据集上的实验表明,提出的方法与传统的ivector系统相比,在EER、minDCF和NIST2011年LRE评价指标上相对提升29.9%-54.6%。
This paper proposes a common service model and its implementation approach for the Sea - Cloud collaborative computing en- vironment. This model abstracts and defines the generic requirements of application development as common services. Targeting at the development of next generation application, the Common Service Platform （CSP） is the fractal -based platform with the feature of open interface and service - orientation. CSP focuses on the integration and interconnection of applications in order to support the intelligent collaboration of＂ People - Computer - Thing＂. CSP aims to implement the sharing, reusable and on - demand mechanism of collabora- tive development for the future co - created ecosystem.
出处
《网络新媒体技术》
2014年第4期40-43,共4页
Network New Media Technology
关键词
音素相关特征
语支鉴别性
因子分析
语支变化量因子
Cloud Computing, Sea Computing, Common Service, Application Development Environment
