The literature suggests that speech representations derived from pre-trained self-supervised systems show similarities to human speech perceptual brain mechanisms, and that subsequent challenges may increase the similarity again by fine-tuning speech positions. Therefore, in this study, we propose to match the model representation with human neural responses using fMRI brain activation to improve the widely applicable wav2vec2.0 model. According to the SUPERB testing analysis, this processing is useful for many underlying tasks such as attention segmentation, automatic voice recognition, and speaker verification. The result is conditioned path recognition the intelligence as an alternative to enhancing self-supervised speech processing.