论文部分内容阅读
为提高噪声不平稳或不可估的情况下语音识别的稳健性,提出了利用自回归模型和短时平稳性假设,估计干净与噪声环境的语音数据,建立相应的语音识别模型,以达到抗噪效果的稳健语音信号处理方法。在N o iseX-92的4种噪声环境(w h ite,babb le,vo lvo,destroyer eng ine)从0到20 dB的不同信噪比下的“863”大词汇连续语音标准数据库的平均识别结果表明,该方法能够使得基于段长分布的隐M arkov模型的语音识别系统在25候选时声学层的音节相对错误率下降达到10.85%以下,同时相对正确识别率上升12.13%。
In order to improve the robustness of speech recognition in case of unsteady or unquantifiable noise, this paper proposes the use of autoregressive model and short-term stability assumption to estimate the speech data of clean and noisy environment and establish the corresponding speech recognition model to achieve anti-noise The effect of robust voice signal processing methods. Average Recognition Results of “863” Vocabulary Consecutive Voice Standards Database with Different Signal-to-Noise Ratio from 0 to 20 dB in 4 Noise Environments of N o iseX-92 (wh ite, babb le, vo lvo, destroyer eng ine) The results show that this method can reduce the syllable relative error rate of acoustical layer below 10.85% while the relative correct recognition rate increases 12.13%.