【摘要】 基于TensorFlow深度学习框架，构建以卷积神经网络为核心的西南官话方言识别系统，研究卷积网络的深度与训练集数据量对系统识别正确率的影响，采用训练损失函数和识别准确率来评估不同深度网络的性能。实验结果表明，相较于浅层和中层网络，深层卷积神经网络对西南官话具有更高的识别正确率。该研究结果可为进一步提升基于深度学习的西南官话方言识别系统的识别效果提供参考。更多还原

【Abstract】 Based on TensorFlow deep learning framework, a southwest mandarin speech recognition system with convolutional neural network as the core is constructed, and the influence of the depth of convolutional network and the data volume of training set on the recognition accuracy of the system is studied. The performance of networks with different depths is evaluated by training loss function and recognition accuracy. The experimental results show that the deep convolutional neural network has a higher recognition accuracy for southwest mandarin than the shallow and middle-level networks. The research results can provide reference for further improving the recognition effect of southwest mandarin recognition system based on deep learning.更多还原