【摘要】 本文在文献[1]建立的外周听觉系统以及部分中枢听觉神经系统的基础上,建立了一个语音识别器。它由听觉模型作为语音声学前端处理器(即特征提取),由具有tonotopic组织结构的神经网络作为识别分类器。大量实验表明,由该听觉模型提取的特征参数不仅能很好地表示语音区别意义,而且对于噪声环境下的语音特征表示有较好的robustness。语音识别实验表明:在有噪声的情况下,采用听觉模型参数的识别器,其识别率明显优于由LPC—倒谱作为语音特征参数的方法。更多还原

【Abstract】 On the basis of the periphery auditory model and partial central auditory neural processing model set up in [1] a speech recognizer using an auditory model as the acoustic front-end preprocessor and a tonotopical organized neural network as the recognition classifier have been built.The experiments show that the parameters derived from the auditory model are a good representation of speech discrimination,especially in noisy environments.The results of speech recognition show that under the condition of 3dB background noise with the same neural network as the classifier,the recognition rate of 3 confusable consonants(p,t,k)is 80.3% for auditory model as the front-end processor and 69.2% for LPC-derived cepstrum as speech parameters ,respectively.更多还原