Page 72 - 201901

第 38 卷 第 1 期 Vol. 38, No. 1
2019 年 1 月 Journal of Applied Acoustics January, 2019

⋄ 研究报告 ⋄

采用骨导语音自适应的语句分割方法 ∗

苗晓孔 † 张雄伟

(陆军工程大学指挥控制工程学院 南京 210007)

摘要 为了解决含噪语句分割问题，也为了解决某些低信噪比环境下传统气导语句分割算法分割效果差、分
割准确度低且算法自适应性弱等问题，提出一种基于骨导语音自适应的分段双门限语音分割方法。将骨导语
音和气导语音同步采集，获取抗噪性能更好的骨导语音，然后在融合过零率与短时能量中引入随机动态阈值
的自适应方法进行端点检测，最后利用分段双门限和语音聚类等手段实现语音分割，提高语音分割算法的鲁
棒性。通过实验验证了所提算法的有效性和可行性，同时与其他语音分割算法进行了对比，证明该文所提分割
算法精度更高，效果更好。
关键词 骨导语音，语音分割，分段双门限，语音聚类
中图法分类号: TP391 文献标识码: A 文章编号: 1000-310X(2019)01-0068-08
DOI: 10.11684/j.issn.1000-310X.2019.01.010

The adaptive speech segmentation method based on bone conduction voice

MIAO Xiaokong ZHANG Xiongwei

(Command & Control Engineering College, Army Engineering University, Nanjing 210007, China)

Abstract In order to solve the problem of segmentation of noisy sentences, and to solve the problems of poor
segmentation efficiency, low segmentation accuracy and poor adaptive ability of traditional air-guided speech
segmentation algorithm in some low SNR environments, a segmentation two-threshold speech segmentation
method based on bone conduction speech adaptation is proposed. Firstly, bone-guided speech and air-guided
speech are acquired synchronously to obtain better anti-noise performance. Then an adaptive method of
random dynamic threshold is introduced to detect endpoints in the fusion of zero-crossing rate and short-term
energy. Finally, segmentation double threshold and speech clustering are used to realize sentence segmentation
and improve the robustness of speech segmentation algorithm. The effectiveness and feasibility of the proposed
algorithm are verified by experiments. At the same time, compared with other speech segmentation algorithms,
the proposed segmentation algorithm is proved to be more accurate and effective.
Key words Bone guided speech, Speech segmentation, Segmented double threshold, Speech clustering

2018-03-18 收稿; 2018-07-13 定稿
国家自然科学基金项目 (61471394)
∗
作者简介: 苗晓孔 (1991- ), 男, 河北石家庄人, 博士, 研究方向: 智能信息处理。
† 通讯作者 E-mail: miao_xk@163.com