This work aims to develop an automatic singing evaluation system for general public. Given a CD/mp3 song recording as the reference basis, the proposed system rates a user's singing performance by comparing it with the vocal in the song recording. This modality allows users to not only enjoy listening to and singing with CD/mp3 songs but also know how well or bad they sing. However, as a majority of songs contain background accompaniments during most or all vocal passages, directly comparing a user's singing performance with the signals in a song recording does not make sense. To tackle this problem, we propose methods to extract pitch-, volume-, and rhythm-based features of the original singer in the accompanied vocals. Our experiment shows that the results of automatic singing evaluation are close to the human rating, where the Pearson product- moment correlation coefficient between them is 0.8. The results are also comparable to those in a previous work using Karaoke music as reference bases, where the latter's task is considered to be easier than that of this work.