In order to enrich students’ vocal learning resources and save teachers’ human resources, this study proposes to build a new virtual teaching platform by combining virtual reality technology with vocal teaching. In the virtual learning system, a feature algorithm for speech spectrograms based on the Log-Gabor Improved Local Binary Pattern (ILBP) is proposed. And a multilevel residual structure CNN is introduced to compensate for the lost features and improve the vocal recognition rate in the vocal teaching process. The results show that the application of the teaching platform proposed in this paper can make the mean value of the post-test scores of the students in the experimental class is an increase of 12.91 points compared with the control class, and the difference between the post-test scores of the students in the two classes is significant (P=0.000). The teaching platform can improve students’ interest in learning vocal music and increase their motivation to learn. Students’ overall satisfaction with the teaching process in the information technology-driven vocal music teaching in music education was 92%. Students’ overall satisfaction with teachers’ choice of teaching media was 91.25%. Satisfaction with the proposed teaching effect in this paper is 95%. It can be seen that the application of the VR technology teaching platform to the music education vocal music teaching classroom has been well received by both students and teachers.