【摘要】 传统语音识别系统中,基于循环神经网络的语音声学模型对长距离历史信息记忆能力有限,难以利用语音的上下文相关性信息,标准长短时记忆单元参数规模庞大,神经网络训练收敛速度较慢。针对以上问题提出一种基于改进门控循环单元的双向循环神经网络的语音识别声学模型。改进模型使用ReLU函数代替双曲正切激活函数,选取单位正交矩阵作为网络初始化参数,结合批量规范化方法,在维持网络长期依赖关系的同时加快训练收敛速度。在TIMIT和LibriSpeech数据集上的实验结果表明:与基线系统相比,改进的门控循环单元模型有2.8%的绝对音素错误率的下降;与标准长短时记忆单元模型相比,神经网络训练的平均迭代周期减少了16.6%,在识别性能和计算效率上均有提升。更多还原

【Abstract】 In the traditional speech recognition system,the speech acoustic model based on the recurrent neural network had limited ability to store long-distance historical information and it was difficult to use the contextual relevance information of the speech. The standard long short-term memory had large scale and the neural network training convergence speed was slow. To solve the above problem,a speech recognition acoustic model based on the bidirectional recurrent neural network with improved gated loop unit was proposed. The ReLU function instead of the hyperbolic tangent activation function was used in improved model;the unit orthogonal matrix was selected as the network initialization parameter. Being combined with the batch normalization method,the training convergence speed was accelerated while maintaining the long-term dependence of the network. Experimental results on the TIMIT and LibriSpeech datasets show that the improved gating recurrent unit model has a 2.8% absolute phoneme error rate reduction compared to the baseline system;being compared to the standard long short-term memory model,the average iteration period of neural network training is reduced by 16.6%,which improves both recognition performance and computational efficiency.更多还原