Reproducibility of the Cepstral Spectral Index of Dysphonia in Tracheoesophageal Speech and Exploratory Associations with Clinical Communication Measures

Background/Objectives: The Cepstral Spectral Index of Dysphonia (CSID) was developed for laryngeal dysphonia. We evaluated task-specific reproducibility in tracheoesophageal (TE) speech and associations with communication measures and surgical/anatomical configuration. Methods: Fifty-four habitual TE speakers were prospectively enrolled. Recordings comprised sustained /a:/ and four connected-speech sentences. Repeated analyses used the same recordings with independent resegmentation. Reproducibility was assessed using intraclass correlation coefficients (ICCs), standard error of measurement, and mean absolute percentage difference (MAPD). The Performance Status Scale for Head and Neck Cancer Patients (PSS-HN) speech subscale and a visual analog scale (VAS) assessed comprehensibility and perceived intelligibility, respectively. PSS-HN agreement used Gwet’s AC2 with ordinal weights. Spearman correlation tests underwent false-discovery-rate adjustment. Configuration-group comparisons used Kruskal–Wallis tests with Benjamini–Hochberg correction and Bonferroni-adjusted pairwise comparisons. Results: Connected-speech single-measure ICCs were 0.987–0.998 within operator and 0.990–0.996 between raters; MAPDs were 1.02–1.99% and 1.41–2.74%, respectively. Sustained-vowel ICCs were 0.749 and 0.874; MAPDs were 8.12% and 5.26%. PSS-HN within-rater and interrater AC2 values were 0.893 and 0.478, respectively. CSID-VAS correlations were modest (absolute ρ = 0.239–0.468). All five tasks differed across the three configuration groups after correction (all q ≤ 0.033). Across all four connected-speech tasks, CSID was lower with cervical–esophageal prosthesis placement after total laryngectomy than with placement into reconstructed tissue. No sustained-vowel pairwise comparison was significant. Conclusions: Connected-speech CSID showed high reproducibility, whereas sustained-vowel measurements showed greater variability. Configuration-group differences cannot isolate an independent effect of prosthesis insertion site. Associations with broader communication measures remain exploratory. Further TE-specific validation against voice-quality ratings remains necessary.