|本期目录/Table of Contents|

基于Multi-Fisher准则的语音混合特征提取和特征增强方法(PDF)

《纳米技术与精密工程》[ISSN:1672-6030/CN:12-1351/O3]

期数:
2017年4期
页码:
317-322
栏目:
精密测量
出版日期:
2017-07-15

文章信息/Info

Title:
Parameter Extraction and Enhancing Method for Mixed Phonetic Features Based on Multi-Fisher Criterion
作者:
赵鑫1 陈晓冬1 常昕1 齐麟2 汪毅1 郁道银1
1. 天津大学精密仪器与光电子工程学院,光电信息技术教育部重点实验室,天津300072; 2. 天津博朗科技有限公司,天津300072
Author(s):
Zhao Xin1 Chen Xiaodong1 Chang Xin1 Qi Lin2 Wang Yi1 Yu Daoyin1
1. Key Laboratory of Opto-Electronics Information Technology of Ministry of Education, School of Precision Instrument and Opto-Electronics Engineering, Tianjin University, Tianjin 300072, China; 2. Tianjin Bolang Technology Co. Ltd, Tianjin 300072, China
关键词:
Multi-Fisher准则 MFCC 语言混合特征 特征增强 嵌入式系统 实时
Keywords:
Multi-Fisher criterion MFCC mixed phonetic features parameter enhancing embedded system real-time
分类号:
TP391.4
DOI:
10.13494/j.npe.20160044
文献标识码:
A
摘要:
提出一种基于多层Fisher(Multi-Fisher)准则的语音混合特征提取和特征增强方法,用于腹腔镜自动定位系统,通过对特定医生的语音指令准确识别,并对非特定人和非特定语音指令拒识,实现对腹腔镜的姿态控制和位置定位,保证腹腔镜位置和姿态的准确性.利用MultiFisher准则对特征参数分类提取和融合,并对特征参数匹配相应的MultiFisher比权重,配合拒识门限,提高特定人语音指令识别精度,增强系统的安全性.实验表明,相比于传统的梅尔频率倒谱系数(MFCC)特征参数,利用Multi-Fisher准则筛选后的特征参数使准确率从86.1%提高到94.2%;经过特征参数增强,准确率进一步提高到95.3%,而识别平均时间仅为283.1 ms,满足实时性要求.
Abstract:
A parameter extraction and enhancing method for mixed phonetic features based on Multi-Fisher criterion is proposed for the embedded automatic positioner for laparoscope (APL). The method can realize the exact location and posture control of APL by recognizing the specific control word from the specific doctor and rejecting different words from different persons, which guarantees the safety and veracity during the period of laparoscope operation. The method extracts and fuses phonetic features, and then it gives the weight based on Multi-Fisher criterion to the fused features, which improves the accuracy and enhances the safety of the APL system with the rejecting threshold. Experimental results show that compared with traditional Melfrequency cepstral coefficients (MFCC), the Multi-Fisher criterion improves the accuracy rate from 86.1% to 94.2%. And the accuracy rate is improved to 95.3% after enhancing the selected phonetic features based on Multi-Fisher criterion, and the average recognition time is just 283.1 ms which meet the real-time requirement.

参考文献/References

-

备注/Memo

备注/Memo:
收稿日期: 2017-03-18. 作者简介: 赵鑫(1991—),男,硕士研究生. 通讯作者: 陈晓冬,教授,xdchen@tju.edu.cn.
更新日期/Last Update: 2017-12-20