摘要
针对归一化方法在连续语音特征曲线调整时存在的问题,提出一种优化解决方案,解决了噪声的不稳定性及不可预测性对语音特征的影响.结果表明,基于该优化方法建立的鲁棒性连续语音识别模型可实现在实验室干净环境和现实噪音环境下同时得到较好的识别结果.
Analyzing the impact of normalization method applied in isolated word speech dominant and noise characteristics to discover the continuous speech characteristic curve adjustment problems. The authors raised optimized solutions to further solve the problem of instability and unpredictability of the noise characteristics for voice effects. Robust continuous speech recognition model by normalization method in this paper can achieve a clean environment in the laboratory and real noise environment so as to get the best recognition results.
出处
《吉林大学学报（理学版）》
CAS
CSCD
北大核心
2015年第3期519-524,共6页
Journal of Jilin University:Science Edition
基金
吉林省自然科学基金(批准号:20140101227JC)
关键词
归一化
噪音鲁棒性
连续语音识别
normalization
noise-robust
continuous speech recognition
