A music training apparatus performs a two-part reproduction including first and second parts, in which the first part is like a trial part which allows a video signal to be reproduced on a display unit but prohibits a sound signal to be generated from a speaker. The second part is like a formal part which allows an ordinary reproduction of both the video and sound signals. In the first part, a first pitch detector detects a pitch of a model singing signal memorized in a recording medium. Meanwhile, a second pitch detector detects a pitch of user's singing signal inputted from a microphone. If the pitches detected by the first and second pitch detectors are different from each other, an interval adjusting device changes the key so as to be in perfect pitch with each other before terminating the first part. Thereafter, in the second part, the reproduction of sound through the speaker formally starts from the head of the music under an ideal condition where the interval is automatically and exactly adjusted to fit to the user's singing sound.