【摘要】 本文讨论了语音识别中样本优化的某些问题。研究了以时间特征序列表示的语音样品用动态规划取平均的方法产生标准参考样本的技术。实现了求取变换函数(动态规划匹配路径)的算法。并对这项技术在发言人及单词识别中的有效性进行了探讨。实验证明这项披术是有效的,明显地改善了识别的精度。更多还原

【Abstract】 This paper reports a method of optimizing the reference templates for the isolated word recognition and speaker recognition.The algorithm is based on linear time warpping and nonlinear time warpping(DP)andaveraging.The algorithm was test in both recognitions mentioned above and thebetter score was obtained.There is evidence that the opt(?)mation approachimproves the performance of recognition system remarkably.更多还原