Expressive synthesis: how crucial is voice quality?

This paper compares the emotive colouring that can be achieved in synthesis by f₀ manipulations alone (f₀ only) as compared to manipulations of f₀ with voice quality (VQ+f₀), and asks how crucial large f₀ excursions are in signalling strong emotions. Are they overwhelmingly important, with voice quality contributing mainly to finer distinctions for milder affects? Or are voice quality and large f₀ differences required for the strong emotions? The ´VQ+f₀´ stimuli, of an utterance synthesised using the LF voice source in KLSYN88 with breathy, whispery, lax-creaky, modal, tense and harsh voice qualities (Gobl et al. (2002)), were further manipulated to replicate the f₀ differences described in Mozziconacci et al. (1995) for 6 emotions, matched to the most appropriate voice quality. The ´f₀ only´ stimuli used the same set of f₀ contours, but retained source settings for modal voice. 10 listeners rated the affective colouring of the stimuli on a seven-point scale, in terms of pairs of opposite attributes. For both strong and milder affects the ´VQ+f₀´ stimuli achieved much higher ratings than the ´f₀ only´ stimuli, which were relatively ineffective. Implications for the synthesis of expressive speech are discussed.