High quality synthesised speech can be produced from the Holmes Parallel-Formant Synthesiser if the control parameters are obtained from analysis of natural speech. However, the quality of the modelling of individual voice characteristics can be very variable. This paper investigates the variability in speech encoded for a parallel-formant synthesiser through the application of objective distortion criteria developed for speech coders. Results are presented for the distortion calculated using the LPC cepstral distance measure.