研究一种基于机器学习算法的音频信号降噪方法(Machine Learning based Audio Denoising Method,MLAADM).该方法综合利用深度神经网络(Deep Neural Network,DNN)、循环神经网络、强化学习技术,实现音频信号降噪.实验结果表明,MLAADM在信噪比、谱失真比、感知评价语音质量及短时客观可懂度等指标上全面优于传统方法和其他深度学习方法,对非平稳噪声处理效果尤为突出,展现了其在复杂噪声环境下的应用潜力.
Research on the Application of Machine Learning Algorithm in Audio Signal Denoising
A Machine Learning based Audio Denoising Method(MLAADM)is studied.The method utilizes Deep Neural Network(DNN),recurrent neural network and reinforcement learning to realize noise reduction of audio signal.The experimental results show that MLAADM is superior to traditional methods and other deep learning methods in signal-to-noise ratio,spectral distortion ratio,perception evaluation of speech quality and short-term objective intelligibility,especially for non-stationary noise processing,demonstrating its application potential in complex noise environments.
NETL
NSTL
万方数据
