SBIR Phase I: Spoken Dialog Question Answering

该小型企业创新研究第一阶段研究项目专注于通过集成自动语音识别(ASR)和最先进的问答(QA)技术来构建当前和下一代对话系统。由于开发这些系统需要非常高的成本、所需的时间以及复杂的设计要求和随之而来的技术专长，口语对话系统的部署有限。该项目的技术目标是：在最少人工干预的情况下自动生成上下文无关语法(CFGs)或统计语言模型(SLM)；自动调整上下文无关语法或统计语言模型以提高ASR转录精度；由于基于QA系统的过滤和反馈而提高转录精度，以及更高水平的语音和语言信息源；查询响应速度高；自然语音输入；对用户输入的准确和有重点的反应。该项目预计，由于减少了人工劳动需求，开发应用程序的成本将显著降低，并且由于存在识别和理解话语所需的更高级别的知识源而提高了准确率。研究项目将由于降低开发成本和提高系统精确度而导致口语对话应用程序的部署显著增加。用户满意度的提高应源于用户输入限制的减少，同时保持较高的目标完成率。此外，在将复杂的任务建模为口语对话系统方面应该更加容易。

This Small Business Innovation Research Phase I research project focuses on building current and next generation dialog systems by integrating automatic speech recognition (ASR) and state-of-the-art question answering (QA) technologies. Because of the very high cost, time required, and the complex design requirements and attendant technical expertise required to develop these systems, spoken dialog systems have had limited deployment. The technical objectives of this project are: automatic generation of context free grammars (CFGs) or statistical language models (SLMs) with minimum manual intervention; automatic tuning of CFGs or SLMs for improved ASR transcription accuracy; higher levels of transcription accuracy due to QA system-based filtering and feedback, and higher level phonetic and linguistic information sources; high levels of query response speed; natural speech input; precise and focused responses to user inputs. This project anticipates a significant reduction in the cost of developing applications due to the reduction of manual labor requirements, and improved accuracy due to the presence of higher level knowledge sources in required for the recognition and understanding of utterances.The research project will result in a significant increase in deployment of spoken dialog applications due to reduced development costs and increased system accuracy. An increase in user satisfaction should follow from the reduction in user input constraints while maintaining high goal completion rates. Additionally, there should be an increased ease in modeling complicated tasks into spoken dialog systems.