The purpose of this study was to investigate what factors affect perceived speaking rate, especially in the case of spontaneous speech. The approach employed in this study was not an experiment in a laboratory condition, but an extraction of the factors from a large scale corpus of spontaneous speech. The materials used were taken from the Corpus of Spontaneous Japanese (CSJ) and its perceived speaking rate database. A subjective speaking rate score on a five step ordinal scale was given to every recording in CSJ. A speaking rate score was a given to each recording as a whole. A canonical discriminant analysis between the perceived speaking rate score and objective attributes extracted from the speech samples was carried out. The results show that the speech of the fastest group had a large number of moras per second, while on the other hand, the speech of the slowest group had a small number of pauses per second, but with a high pause ratio.
