Abstract
Nonconformity reporting and measurement traceability are core practices in quality management, yet their effectiveness is frequently limited by the speed and accuracy of shop-floor data capture. In apparel and fashion supply chains, where inspections combine structured measurements with unstructured notes, operators often alternate between manual transcription, spreadsheets, and heterogeneous terminology. This paper presents a speech-enabled Quality Control (QC) data capture system that converts spoken inspections into structured, database-ready records. The proposed artifact integrates an audio ingestion module with automatic duration detection and segmentation, a cloud-based Voice-to-Text transcription, information extraction based on a Large Language Model and constrained by domain dictionaries, and a human-in-the-loop verification interface before final persistence to a relational database. The system manages vocabulary governance as it detects out-of-dictionary measurement points and defect terms, logs them as “unrecognized,” and provides an operator workflow to extend dictionaries by adding new normalized elements or synonyms. The paper describes the architecture, data model, prompt-driven extraction strategy, and vocabulary-governance mechanisms, and reports a prototype-level technical evaluation on annotated QC inspection audio recordings. The contribution is a deployable approach designed to support traceable, standardized, and analysis-ready QC data capture with human validation before final persistence.