摘要
近年来,孤立词语音识别技术由于其对计算量存储量要求低和灵活性高的特性而备受关注。但由于当今生活环境的复杂性和不确定性,使得孤立词语音识别技术在实时性和准确性方面仍面临着巨大的挑战。为此,以非特定人孤立词语音识别为研究对象,将改进的过门限率、子带谱熵及Teager能量算子(TED)相融合进行语音的端点检测,以找出合理的语音起始点,并将其应用在基于隐马尔可夫模型的语音识别系统中,通过直观的语音识别正确率来验证该方法的优越性。通过实验仿真,与其他的传统方法进行对比,所提方法可使得语音识别系统满足一定的实时性要求,且在孤立词识别的准确性和稳定性上占一定优势。
In recent years, speech recognition technology has been highly developed and widely used, Among them, the isolated-word speech recognition technology has attracted much attention due to its low requirements and high flexibility for the amount of computational storage. However, the isolated-word speech recognition still faces great challenges both in real-time recognize and accuracy. This paper takes the non-specific human isolated-words speech recognition as the research object, combines Subband spectral entropy with Teager energy operator to detect the endpoint of speech signals, and applied it to speech recognition system based on hidden markov model. The experiment results shows that the proposed method can meet the real-time recognize requirements, and has a better effect of isolated-words speech recognition in accuracy rate and stability.
作者
卢洵波
李昕
Lu Xunbo;Li Xin(School of Mechatronic Engineering and Automation,Shanghai University,Shanghai 200444,China)
出处
《电子测量技术》
2020年第7期129-136,共8页
Electronic Measurement Technology
关键词
孤立词识别
子带谱熵
Teager能熵比
过门限率
isolated-word speech recognition
subband spectral entropy
Teager energy operator
threshod-crossing rate
