论文部分内容阅读
针对鲁棒语音识别中的声效模式检测问题,提出了一种分级检测方法.首先使用整体谱特征训练高斯混合模型来判定语音信号是否耳语.对于非耳语的语音信号,通过声学界标点检测来获取信号中的元音段,然后通过元音模板匹配来确定语音信号具体的声效模式.在863-test测试集上进行的声效检测实验结果显示,除耳语识别精度略有下降外,其他4种声效模式的识别精度均有大幅度的提高.实验结果表明了将语音信号整体特征与局部元音特征相结合在声效检测中的有效性.
Aiming at the problem of sound mode detection in robust speech recognition, a hierarchical detection method is proposed. First, Gaussian mixture model is used to train the Gaussian mixture model to determine whether the speech signal is whisper. For non-whisper speech signals, acoustic landmark punctuation detection Signal vowel segments, and then through the vowel template matching to determine the specific sound signal mode of the sound effect in the 863-test test set sound effects on the experimental results show that, in addition to a slight decline in the recognition accuracy of the whisper, the other four sound effects The recognition accuracy of the model has been greatly improved.The experimental results show the effectiveness of combining the overall characteristics of the speech signal with the local vowel features in the sound detection.