基于多核学习的多带抗噪声语音识别方法仿真

Multi-band Anti-Noise Speech Recognition Method Simulation Based on Multi-Core Learning

由于传统语音识别方法在安静环境下语音识别较为准确,可在现实环境下,噪声干扰语音特征提取,导致测量数据不可信,语音识别方法正确率低.提出一种基于多核学习的多带抗噪声语音识别方法,构建多核学习组合算法.算法是多核学习与投影算法的融合,根据不同频带带宽,可以将多带噪声有效地分类,并加强语音特征级,与CHMM模型共同完成多带抗噪声语音识别方法,计算得出各模型条件概率,数值最大的即是语音识别结果.根据仿真结果分析,基于多核学习的多带抗噪声语音识别方法,相比传统方法可以减少运算量,提高语音识别正确率,有效地识别出现实噪声环境下正常语音.

In quiet environment, the traditional speech recognition method is more accurate. In real environment, the speech feature extraction is influenced by noise interference, resulting in unreliable measurement data and low accuracy of speech recognition. Therefore, this article proposed a multi-band anti-noise speech recognition method based on multi-core learning. Firstly, the multi-core learning combination algorithm was constructed. This algorithm is the fusion of multi-core learning and projection algorithm. According to different bandwidths, the multi-band noise could be classified effectively and the speech feature level could be strengthened. After that, the multi-band anti-noise speech recognition system was completed with CHMM model. Then, conditional probability of each model was calculated. Thus, the maximum value was the speech recognition result. According to the analysis of simulation result, the multi-band anti-noise speech recognition method based on multi-core learning can effectively reduce the computational complexity and improve the accuracy of speech recognition. Thus, this method can effectively recognize the normal speech in real noise environment.