面向变电站智能运检的声音谱特征语音识别方法

Speech Recognition Method of Sound Spectrum Feature for Intelligent Operation and Maintenance of Substations

语音识别是变电站智能运检中关键的人机交互技术。然而,由于生产环境中存在使用专业术语多和噪声大的问题,传统的语音识别方法的效果受限。为此,文中提出了一种基于声音谱特征的语音识别方法。通过融合MFCC与CQT谱,形成一种基于声音谱的特征参数,通过对参数分布的估计,能够有效地降低语音信息中的噪声干扰。为提升语音识别性能,文中设计一个端到端的语音识别模型。该模型基于卷积神经网络(CNN),并融合了CTC和注意力机制。CNN网络能够有效地捕捉语音数据中的局部模式和结构信息,而CTC和注意力机制在解码过程中起到关键作用。文中使用Aurora、Aishell以及运检语音数据集进行了实验评估,比较了语音降噪、语音识别同传统方法的效果。实验结果表明,所提出的语音识别模型取得了显著的性能提升,可为相关领域的研究和应用提供有价值的参考。

Speech recognition is a key technology for human-computer interaction in intelligent operation and maintenance of substations.However,the effect of traditional speech recognition method is limited due to the presence of numerous technical terms and loud noise.Therefore,a speech recognition method based on sound spectral features is proposed in this paper.By fusing MFCC(Mel-frequency ceptral coefficients)and CQT(constant-Q transform)spectra,a spectral-based feature parameter is formed,which effectively reduces noise interference in the speech information through estimation of parameter distribution.For improving the performance of speech recognition,an end-to-end speech recognition model is designed in this paper.The model is based on Convolutional Neural Networks(CNN)and fusion of the CTC(connectionist temporal classification)and attention mechanisms.The CNN network can be able to effectively capture local patterns and structural information in speech data,while the CTC and the attention mechanism play a crucial role in the decoding process.The Aurora、Aishell and operation and maintenance speech data set are used in this paper for experimental assessment.The effect of speech noise reduction and speech recognition is compared with the traditional method.The experimental results show that the proposed speech recognition model performs well in performance enhancement and provides valuable references for research and applications in related fields.