Scholarlink

This work presents a new AI-based approach that employs a single, effective pipeline to convert speech into SQL queries. Our approach integrates two state-of-the-art models, T5 for text-to-SQL translation and Wave2Vec2 for speech recognition, into one system that eliminates the need for multi-step processes. Unlike conventional methods that treat text-to-SQL and speech-to-text as isolated tasks, our model directly interprets voice instructions into formal SQL queries. This integration improves the efficiency of processing, reduces latency, and eliminates intermediate step errors. The model can also dynamically adapt to speech differences, voices, and domain-specific jargon due to a feedback loop system that allows users to review and refine generated queries. Through natural language interaction with data analytics dashboards, the proposed approach brings technical and non-technical users closer to their databases and natural language. To compare our results from testing to conventional approaches, we demonstrate better accuracy, faster query development, and an improved user interface. This research presents the possibility for more efficient and accessible data-driven decision-making via AI and NLP technology.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Speech-To-SQL: Transformer-Based Models for Interactive Dashboarding

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Aastha Anand,
  • Harsh Patel,
  • Yogiraj Anil Bhale

摘要

This work presents a new AI-based approach that employs a single, effective pipeline to convert speech into SQL queries. Our approach integrates two state-of-the-art models, T5 for text-to-SQL translation and Wave2Vec2 for speech recognition, into one system that eliminates the need for multi-step processes. Unlike conventional methods that treat text-to-SQL and speech-to-text as isolated tasks, our model directly interprets voice instructions into formal SQL queries. This integration improves the efficiency of processing, reduces latency, and eliminates intermediate step errors. The model can also dynamically adapt to speech differences, voices, and domain-specific jargon due to a feedback loop system that allows users to review and refine generated queries. Through natural language interaction with data analytics dashboards, the proposed approach brings technical and non-technical users closer to their databases and natural language. To compare our results from testing to conventional approaches, we demonstrate better accuracy, faster query development, and an improved user interface. This research presents the possibility for more efficient and accessible data-driven decision-making via AI and NLP technology.