基于信号处理的电话语音模拟  

Telephone Speech Simulation Using Signal Processing Method

作　　者：左国玉 刘文举[1] 阮晓钢[2] 

机构地区：[1]中国科学院自动化所模式识别国家重点实验室,北京100080 [2]北京工业大学电子信息与控制工程学院,北京100022

出　　处：《北京工业大学学报》2003年第2期182-187,共6页Journal of Beijing University of Technology

基　　金：国家自然科学基金(60172055);中国科学院自动化所领域前沿基金(1M02J05)

摘　　要：针对电话语料比较缺乏的问题,提出了一种完全由软件模拟实现由纯净语音向电话质量语音转换的算法。该算法采用滤波器设计技术来模拟电话线路连接中各种模拟传输设备频率响应特性,并对电话通道环境中各种噪音行为等电话语音现象进行了模拟研究。频谱失真分析和识别实验结果表明,通过模型参数的合理设置和调整,该算法能有效地实现纯净语音向电话质量语音的近似模拟,使用由纯净数据生成的模拟语音可以获得与真实语音同等的识别性能。Owing to the lack of telephone speech data, this paper proposes a software simulation implementation of converting clean speech sounds into telephone-quality ones. Filter design technology is adopted to simulate the frequency response characteristics of various analogue transmission equipments in telephone circuit connection and to make simulation study on such telephone speech phenomena as different noise behaviours in telephone channels. The spectral distortion analysis and recognition experiment results show that through the reasonable setting and regulating, the algorithm can effectively realize the approximate simulation from clean speech to telephone-quality ones so that the simulated speech sounds generated from clean data can achieve as good recognition performance as real speech.

关 键 词：电话语音模拟 信号处理 滤波器 频谱失真 语音识别 

分 类 号：TN912.34[电子电信—通信与信息系统]