With the digitalization of vocational education, intelligent pronunciation training has been introduced as a new form in the reform of English teaching at vocational colleges. Traditional phonetics teaching has issues, such as delayed feedback, a small training volume, subjective evaluation, and a lack of personalized guidance. To investigate the actual effects of AI on vocational students’ second-language pronunciation acquisition, this study selects 42 students majoring in primary school English Education as subjects and employs a one-group pretest-posttest design for a one-semester AI teaching intervention. Praat, professional acoustic software, was used to conduct quantitative comparisons of the pre- and post-intervention speech data at various levels. Through the research, AI-assisted teaching has helped students improve the acoustic features of their pronunciation, correct typical errors, and enhance monotonous intonation and fragmented speech; meanwhile, the accuracy and fluency of their speaking have been significantly improved. Based on objective acoustic data, this study has verified the practical value of AI teaching and reduced the range of subjective evaluation in traditional phonetics classes. It has established a closed-loop teaching model of daily AI training and regular Praat acoustic tests, and provided replicable empirical schemes for the digital reform, quantitative pronunciation evaluation, and personalized phonetic teaching in vocational colleges.