论文部分内容阅读
基音周期提取是语音编码和语音识别领域的一项重要研究课题。为了解决传统的自相关方法容易出现的半频倍频错误,提出了基于时域和频域分析的提取算法。该算法首先提取时域自相关值最大的若干个候选值;然后统计每个候选值对应的频域上所有相邻两个谐波能量和的最大值,用来对其自相关值进行加权;最后根据历史的基音周期值以及候选基音周期所对应的频域能量值对加权值进行修正。使用Keele数据库进行测试表明,使用该算法后基音周期提取的半频倍频错误率比传统算法下降了50%左右。
Pitch extraction is an important research topic in the field of speech coding and speech recognition. In order to solve the half-frequency doubling error prone to the traditional autocorrelation method, an extraction algorithm based on time-domain and frequency-domain analysis is proposed. Firstly, the algorithm extracts several candidate values with the largest autocorrelation value in the time domain. Then, the maximum value of the sum of the energies of all the two adjacent harmonics in the frequency domain corresponding to each candidate value is calculated, and the autocorrelation value is weighted. Finally, the weighted value is corrected according to the historical pitch period value and the frequency domain energy value corresponding to the candidate pitch period. The test using Keele database shows that the half-frequency doubling error rate of pitch extraction using this algorithm has dropped by about 50% compared with the traditional algorithm.