Several experiments have suggested that deliberate changes of the speaker's voice quality result in modifying the LTAS of his speech productions. The few studies in this field have mainly dealt with speaker recognition tasks under various conditions, where the speakers were simply asked to modify their voices. However, Nolan (19$3) had 2 trained subjects utter 30 times an invariant text, using a given 'setting' in each utterance. He computed a LTAS for each utterance and mainly performed a qualitative study of the spectral traces. The purpose of our experiment is to combine Nolan's approach, allowing to control the introduced vocal variability, with a quantitative technique allowing to measure the resulting spectral variability. One trained subject has produced a French balanced text with 31 different qualities. Each utterance has been analysed by means of a 2033 Bruel Kjaer FFT analyser. The so obtained LTAS were compared by means of the SDDD dissimilarity index. The results show what groups of settings deliver similar spectral results.