Electronic music systems can achieve intelligent control of musical instruments, but there are still certain difficulties in extracting vocal features, especially when vocal singing is affected by various factors, resulting in a high error rate in extracting the main features of vocal singing. This article uses a deep neural network fusion method to extract the main features of vocal singing. The purpose is to analyze the waveform of vocal singing through intelligent methods and improve the extraction effect of the main features of vocal singing. In addition, this paper studies and validates time-frequency denoising and denoising, and proposes a new denoising threshold from the aspects of complexity and engineering, which can be better applied to engineering and time-frequency joint denoising. After the denoising is completed, this paper compares the time-frequency map for noise removal and the morphological filter for noise removal. Furthermore, given the computational complexity and completeness of extraction, this paper proposes a method to remove interference by applying connected-domain labeling. The quantitative evaluation results of the main feature extraction method in vocal singing based on deep neural networks are distributed between [87 , 92], indicating that the proposed method can achieve high accuracy in main feature extraction in vocal singing, the simulation results show that the main feature extraction method for vocal singing based on deep neural networks proposed in this paper is feasible, and it can also play an important role in the main feature extraction of vocal singing. A method of using connected domain labeling to remove interference is proposed, which can be combined with existing music waveform algorithms for fusion, facilitating the practical application of subsequent intelligent models. The method proposed in this article effectively promotes the intelligent application of vocal feature recognition systems.