Humming training can help students overcome problems in areas such as pitch and interval difficulty and improve their musical skills. In this paper, we construct a humming training recognition model using a transfer learning algorithm as a way to improve vocal skills. The study adopts the convolutional network VGG-16 as the base model and fine-tunes it to obtain the model IVGG.The convolutional block of the IVGG model is used as a feature extractor to extract features from the preprocessed hum training corpus, and the effect of hum training is detected according to the model output. This paper’s method has a recognition rate of 81.33%. Based on the method of this paper, vocal skill enhancement training is conducted and compared with the traditional training model. The difference of students’ scores on the three dimensions of vocal learning interest, vocal learning attitude, and vocal learning ability are 0.508, 0.493, and 0.391, respectively, and the p-values are 0.003, 0.000, and 0.003, respectively.Compared with the traditional humming training mode, the humming training recognition model based on the transfer learning algorithm can be used to effectively improve the vocal music skills.