摘要
连续语音识别技术是人工智能领域的研究热点之一,连续语音切分是语音识别的关键环节。实现准确可靠的连续语音切分算法,能够明显提升连续语音识别系统的性能,符合目前连续语音识别系统的应用需求。论文通过分析汉语语音识别的建模单元、汉语的语言结构和发音规律,综合利用语音信号的时域特征、频域特征与倒谱域特征,依据基音周期轨迹的断点和斜率变化,设计浊音检测与多级切分算法,实现了连续汉语语音切分技术。实验表明,算法能够取得较好结果,且在噪声环境下具有较好的鲁棒性。
Continuous speech recognition technology is one of the research hotspots in the field of artificial intelligence.Con⁃tinuous speech segmentation is the key link of speech recognition.The realization of accurate and reliable continuous speech segmen⁃tation algorithm can significantly improve the performance of continuous speech recognition system and meet the application require⁃ments of continuous speech recognition system at present.In this paper,by analyzing the modeling of Chinese speech recognition unit,the structure of the Chinese language and pronunciation rules,time domain,frequency domain and cepstrum characteristics of speech signal are used,dullness detection and multi-level segmentation algorithm are designed by analysing the break point and slope change of the pitch periodic track,then the continuous Chinese speech segmentation method is realized.Experiments show that the algorithm is robust under noise environment.
作者
高桥
张二华
GAO Qiao;ZHANG Erhua(School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094)
出处
《计算机与数字工程》
2023年第1期163-167,共5页
Computer & Digital Engineering
关键词
语谱图
倒谱
基音周期谱
基音周期轨迹
多尺度分析
speech spectrum
cepstrum
pitch periodic spectrum
pitch periodic track
multiscale analysis
