Modeling intelligibility deficits in dysphonic speech: Toward development of a Dysphonic Speech Transmission Index (DSTI)

Dysphonia degrades not only voice quality but also speech intelligibility, particularly in noisy environments. To date, there are no established acoustic models for describing intelligibility deficits associated with dysphonia. Amplitude modulation (AM) patterns, which encode critical prosodic and rhythmic information, may be especially vulnerable to degradation in dysphonic speech. This pilot study examined whether reductions in AM strength, derived from modulation transfer function (MTF) analysis, predict intelligibility loss in dysphonic speech and explored the feasibility of a Dysphonic Speech Transmission Index (DSTI). Audio recordings of 6 dysphonic and 6 vocally healthy speakers reading standardized sentences were analyzed. MTF coefficients were extracted across 7 octave bands (125 Hz–8 kHz) and 14 modulation frequencies (0.63–12.5 Hz). A LASSO regression model was used to predict intelligibility, indexed by listener-derived word error rates. Model performance and feature selection were assessed using cross-validation. The LASSO model showed strong performance (94.5% accuracy; MSE = 0.055). Key predictors clustered in the 1–4 Hz modulation range, supporting the importance of slow AM cues for perception. AM disruptions contribute to intelligibility deficits in dysphonic speech. These findings support the feasibility of DSTI as a scalable, physiologically grounded intelligibility model.