论文部分内容阅读
提出了一种新的用于语音识别的HMM MLP混合网络 ,它利用MLP的鉴别训练能力 ,以克服基本HMM的ML训练方法中不合理的模型正确性假设前提 ,提高HMM的鉴别能力和识别性能 .实验结果证明HMM MLP混合网络的鉴别能力和识别性能明显高于普通HMM .
A new HMM MLP hybrid network for speech recognition is proposed, which utilizes MLP discriminating training ability to overcome the unreasonable assumptions of model correctness in ML training method of basic HMM and improve the discriminating ability and recognition performance of HMM Experimental results show that the discriminative ability and recognition performance of HMM MLP hybrid network are obviously higher than that of normal HMM.