Year: 2026
Type: article
Abstract: This paper presents a comprehensive approach to database modeling for software systems designed to support communication for hearing- and speech-impaired users. The proposed model integrates Speech-to-Text (STT), Neural Machine Translation (NMT), and Text-to-Speech (TTS) components within a unified relational schema based on the IDEF1X methodology. The database structure ensures logical consistency, scalability, and efficient data processing across all functional modules. Each component maintains its core entities while sharing common relational tables for languages, neural models, evaluation metrics, and version control. The STT module models the flow from audio input to segmented transcription and quality evaluation using WER and CER indicators. The NMT module structures translation processes from source text and tokenization to translated output and BLEU-based assessment. The TTS module represents text segmentation, synthesized audio generation, prosodic features, and naturalness evaluation. The integrated model enables systematic monitoring, comparison of model versions, and performance analysis at segment, sentence, and token levels. This unified database architecture supports end-to-end processing (speech–text–translation–speech), improves data integrity, and provides a scalable foundation for further optimization and intelligent system development.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7161481410