论文部分内容阅读
提出一种基于自组织聚类,并且利用改进粒子群算法确定转换模型参数的语音转换方法.该方法首先基于自组织特征映射网络对特征参数进行聚类,再对每个聚类分别建立转换规则,并且利用柯西变异的粒子群算法确定每个转换规则中的模型参数.与传统的单一转换规则相比,聚类后建立的多转换规则以及利用改进粒子群算法确定参数能够提高映射关系的准确度,避免参数陷入局部最优点。以女声转男声为例,主观测试表明该方法得到的转换语音与目标的相似度提高了27.6%,平均主观意见分(Mean Opinion Score,MOS)提高了0.6,客观测试也表明该方法谱失真最小,与目标的包络更接近.
This paper proposes a speech transformation method based on self-organizing clustering and using the improved particle swarm optimization algorithm to determine the parameters of the transformation model.The method firstly clusters the feature parameters based on the self-organizing feature mapping network and then establishes the conversion rules for each cluster , And the Cauchy mutation particle swarm optimization algorithm is used to determine the model parameters of each transformation rule.Compared with the traditional single transformation rules, the multi-transformation rules established after clustering and the use of improved particle swarm optimization to determine the parameters can improve the mapping relationship Accuracy, to avoid the parameters fall into the local optimum. Taking male voice as male voice as an example, the subjective tests show that the similarity between converted voice and target obtained by this method is improved by 27.6% and Mean Opinion Score (MOS) is improved by 0.6. Objective tests also show that this method has the lowest spectral distortion , Closer to the target’s envelope.