摘要
针对传统的基于码书映射的语音转换算法存在的与目标语音相似度不高和高频段细节信息丢失等问题,提出基于码书映射的语音转换方法改进方案。通过提出基于转换权重预测重构的码书映射关系,对语音特征参数的权重进行转换,提高转换结果与目标语音特征参数的相似度。通过建立残差数据库并对转换后声道参数的高频率段进行残差补偿,降低高频率段的语音失真。实验结果表明,与传统的基于码书映射的语音转换方法相比,改进后的转换方法可有效提高语音转换效果。
The traditional codebook conversion algorism has some problems such as low similarity to target voice and the loss of high- frequency section detail information. So the voice improved schemes based on codebook mapping are put forward in this paper. Firstly,a codebook mapping relationship based on conversion weight prediction refactoring is provided to promote the similarity between the conversion result and target voice characteristic parameters. The simulation results show that the improved conversion method shows better conversion effects than the traditional codebook mapping method.
出处
《微处理机》
2015年第2期35-38,43,共5页
Microprocessors
基金
中央高校基金(2012B04014)
教育部重点实验室开放基金(NYKL201305)
关键词
语音转换
码书映射
权重转换
残差补偿
Voice Conversion
Codebook Mapping
Codebook Weight Conversion
Residua Compensation
