There is growing interest in using digital health technologies to overcome limitations of traditional negative symptom assessment in schizophrenia spectrum disorders (SSD). For example, computerized speech analysis has been used to quantify speech and language changes associated with negative symptom severity in SSD. However, the large number of speech features and variability across studies make it difficult to determine the features with the greatest clinical utility. The present study aimed to identify the most robust speech-based markers of negative symptom severity by comprehensively evaluating their psychometric properties in a sample of individuals with SSD. Data were analyzed from 62 SSD participants who completed clinical assessments and speech tasks at baseline and follow-up visits. Thirty-eight acoustic and linguistic speech features were evaluated for test-retest reliability within visits, associations with clinician-rated negative symptoms, convergent and discriminant validity, specificity for negative symptom severity, and associations with participant clinical and demographic characteristics. Consistency of findings was evaluated across the two study visits, treating the second visit as a within-study replication sample. Three features (speech proportion, speech rate, unfilled pauses) consistently showed adequate or better reliability, correlations with negative symptoms across tasks and visits, and further demonstrated discriminant validity, specificity, and lack of associations with antipsychotic medication and extrapyramidal symptoms. Speech rate also consistently demonstrated convergent validity with an alternative negative symptom measure and was the most reliable feature when evaluated at the level of a single task administration. These findings advance the clinical validation of speech-based digital biomarkers for negative symptom assessment in SSD.