OBJECTIVE : To analyze the effects of auditory-perceptual training with anchor stimuli of natural voices on inter-rater agreement during the assessment of vocal quality. METHODS : This is a quantitative nature study. METHODS : An auditory-perceptual training site was developed consisting of Programming Interface A, an auditory training activity, and Programming Interface B, a control activity. Each interface had three stages: pre-training/pre-interval evaluation, training/interval, and post-training/post-interval evaluation. Two experienced evaluators classified 381 voices according to the GRBASI scale (G-grade, R-roughness, B-breathiness, A-asthenia, S-strain, I-instability). Voices were selected that received the same evaluation by both evaluators: 57 voices for evaluation and 56 for training were selected, with varying degrees of deviation across parameters. Fifteen inexperienced evaluators were then selected. In the pre-, post-training, pre-, and postinterval stages, evaluators listened to the voices and classified them via the GRBASI scale. In the stage interval evaluators read a text. In the stage training each parameter was trained separately. Evaluators analyzed the degrees of deviation of the GRBASI parameters based on anchor stimuli, and could only advance after correctly classifying the voices. To quantify inter-rater agreement and provide statistical analyses, the AC1 coefficient, confidence intervals, and percentage variation of agreement were employed. RESULTS : Except for the asthenia parameter, decreased agreement was observed in the control condition. Improved agreement was observed with auditory training, but this improvement did not achieve statistical significance. CONCLUSIONS : Training with natural voice anchors suggest an increased inter-rater agreement during perceptual voice analysis, potentially indicating that new internal references were established.