Our goal in this review is to map how voice diaries are used as digital biomarkers, summarize computational pipelines, and identify methodological gaps. Seven databases were searched using the Population-Concept-Context framework. 10/41 records met inclusion criteria. Most studies used transcripts (7/10); few analyzed audio (3/10), one used a multimodal approach with text, audio, and sensor data. Methods spanned lexicon-based, machine learning, and LLM approaches. Validation was inconsistent. Computational analysis of voice diaries is promising but early-stage and text-based requiring transparent methods and rigorous validation.