摘要
针对永嘉方言的口音及音调的特点,本文提出一种基于改进黑翅鸢优化算法的LSTM语音识别模型。首先,针对传统黑翅鸢优化算法的缺陷,引入Chebyshev混沌序列优化黑翅鸢种群初始化流程;并在位置更新过程中,结合透镜反向学习策略平衡算法的全局寻优与局部勘探能力;提高算法的收敛精度和收敛速度,为防止算法早熟收敛,通过加入改进高斯变异因子,以引入适当扰动的方式帮助算法跳出局部最优。其次,将改进后的黑翅鸢算法与LSTM网络结合,搭建IBKA-LSTM语音识别模型。最后,通过梅尔倒谱系数对特征方言进行提取,并通过IBKA-LSTM模型进行识别。
Aiming at the accent and tonal characteristics of the Yongjia dialect,this paper proposes a Long Short-Term Memory(LSTM)speech recognition model based on an Improved Black-winged Kite Optimization Algorithm(IBKA).Firstly,to address the shortcomings of the traditional Black-winged Kite Optimization Algorithm(BKA),the improved algorithm introduces the Chebyshev chaotic sequence to optimize its population initialization process.During the position update phase,it incorporates a lens oppositionbased learning strategy to balance the algorithm's global exploration and local exploitation capabilities,thereby enhancing its convergence accuracy and speed.To prevent premature convergence,an improved Gaussian mutation factor is added to introduce appropriate perturbations,helping the algorithm escape from local optima.Secondly,the improved Black-winged Kite Algorithm is integrated with the LSTM network to construct an IBKA-LSTM speech recognition model.Finally,Mel-Frequency Cepstral Coefficients(MFCC)are used to extract features from the characteristic dialect speech,and recognition is performed by the IBKALSTM model.
作者
张志强
杨振梅
Zhang Zhiqiang;Yang Zhenmei(School of Artificial Intelligence,Zhejiang Dongfang Polytechnic,Wenzhou,Zhejiang 325000,China;Wenzhou Polytechnic)
出处
《计算机时代》
2025年第10期44-50,共7页
Computer Era
基金
2024年度温州市科协服务科技创新项目:基于ResCNN-BiGRU的永嘉方言语音识别技术研究(项目编号:KJFW2024-038)。
