The talker dependency of the correct response rate on the English/r/-/l/ perception by Japanese listeners has been reported. This paper reports that the acoustical dissimilarity of an/r/-/l/ minimal pair measured by a dynamic cepstrum shows a talker dependency pattern closely correlated with the talker dependency pattern of the correct response rate. Nine spectral representations are compared in measuring the /r/-/l/ acoustical dissimilarity. The nine spectral representations include three base parameters, a cepstrum, a weighted cepstrum, and a mel-cepstrum, and dynamic cepstra and delta cepstra. derived from the three base parameters. The experimental results imply that Japanese listeners tend to identify /r/ or /l/ using the succeeding vowels affected by coarticulations.