Perceptual equivalence of the Liljencrants-Fant and linear-filter glottal flow models

Speech glottal flow has been predominantly described in the time-domain in past decades, the Liljencrants-Fant (LF) model being the most widely used in speech analysis and synthesis, despite its computational complexity. The causal/anti-causal linear model (LF_CALM) was later introduced as a digital filter implementation of LF, a mixed-phase spectral model including both anti-causal and causal filters to model the vocal-fold open and closed phases, respectively. To further simplify computation, a causal linear model (LF_LM) describes the glottal flow with a fully causal set of filters. After expressing these three models under a single analytic formulation, we assessed here their perceptual consistency, when driven by a single parameter R_d related to voice quality. All possible paired combinations of signals generated using six R_d levels for each model were presented to subjects who were asked whether the two signals in each pair differed. Model pairs LF_LM-LF_CALM were judged similar when sharing the same R_d value, and LF was considered the same as LF_LM and LF_CALM given a consistent shift in R_d. Overall, the similarity between these models encourages the use of the simpler and more computationally efficient models LF_CALM and LF_LM in speech synthesis applications.