The author proposes a model, DRSC, that disentangles intent information from text and Mel-spectrogram domains to improve medical speech symptoms classification accuracy.