传统声学特征和非线性特征用于病态嗓音的比较研究

发布时间：2018-03-01 13:04

本文关键词： 病态嗓音支持向量机传统声学特征非线性动力学技术　出处：《生物医学工程学杂志》2014年05期 　论文类型：期刊论文

【摘要】：本文通过分析嗓音的发音机制,提取正常与病态嗓音的传统声学参数[基频、Mel倒谱系数(MFCC)、线性预测系数(LPCC)、频率微扰、振幅微扰]与非线性动力学特征参数[熵(样本熵、模糊熵、多尺度熵)、计盒维数、计维截距和Hurst参数],作为病态嗓音识别的特征矢量集。应用支持向量机(SVM)对/a/音的78例正常嗓音与73例病态嗓音和/i/音的78例正常嗓音与80例病态嗓音进行建模与识别。结果表明,相对于传统的声学特征参数,非线性特征参数能更好地区分正常与病态嗓音;实验提取的所有参数中,除了多尺度熵,/a/音的正常与病态嗓音的识别率均高于/i/音,因此为了达到识别病态嗓音的目的,国内外相关研究大多采用/a/音数据;多尺度熵特征对/i/音的正常与病态嗓音的识别率较/a/音高,它或能为评价声带代偿功能状态的研究提供有益的启发。
[Abstract]:By analyzing the phonetic mechanism of voice, the traditional acoustic parameters (fundamental frequency Mel cepstrum coefficient MFCCP, linear prediction coefficient LPCCX, frequency perturbation, amplitude perturbation) and nonlinear dynamic characteristic parameter [entropy, sample entropy,] are extracted. Fuzzy entropy, multi-scale entropy, box-counting dimension, Dimension intercept and Hurst parameter], as feature vector set for pathological voice recognition. Using support Vector Machine (SVM) to model 78 normal voices of / R / A, 78 normal voices of 73 pathological voices and 78 normal voices and 80 sick voices of / R / A voice. And recognition. The results show that, Compared with the traditional acoustic characteristic parameters, the nonlinear characteristic parameters can better distinguish the normal voice from the pathological voice, except for the multi-scale entropy / a / sound, the recognition rate of the normal voice and the pathological voice is higher than that of the / I / sound. Therefore, in order to identify the pathological voice, most of the related studies at home and abroad adopt the / R / A / sound data, and the recognition rate of the normal / pathological voice of the / I / sound is higher than that of the / R / R, It may provide useful inspiration for the evaluation of vocal compensatory functional state.
【作者单位】：广西师范大学电子工程学院;
【基金】：国家自然科学基金资助项目(61362003) 广西自然科学基金资助项目(2010GXNSFA013128)
【分类号】：R312

【参考文献】