project overview
play

Project Overview Speech Speech Generation Generation Common - PowerPoint PPT Presentation

9807-11 Multilingual Conversational System Research James Glass and Stephanie Seneff Project Overview Speech Speech Generation Generation Common Semantic Frame Speech Speech Understanding Understanding DATABASE Explore


  1. 9807-11 Multilingual Conversational System Research James Glass and Stephanie Seneff Project Overview Speech Speech Generation Generation Common Semantic Frame Speech Speech Understanding Understanding DATABASE • Explore language-independent approaches to speech understanding and generation • Develop necessary human-language technologies to enable porting of conversational interfaces from English to Japanese • Use existing Jupiter weather-information domain as test case NTT - MIT Research Collaboration — Bi-Annual Report, January 1—June 30, 2000

  2. 9807-11 Multilingual Conversational System Research James Glass and Stephanie Seneff Progress Through June 2000 • Developed prototype weather information system (Mokusei) using MIT human language technology • Initiated data collection from NTT employees – Appropriate answers to approximately 60% of queries • Ongoing system refinement: – Upgrade of language generation component for more natural sounding Japanese output – Reduce vocabulary inconsistencies among recognizer, parser, database, and generation components NTT - MIT Research Collaboration — Bi-Annual Report, January 1—June 30, 2000

  3. 9807-11 Multilingual Conversational System Research James Glass and Stephanie Seneff Research Plan for the Next Six Months • Collect spontaneous speech data from more speakers interacting with the Mokusei system • Develop natural sounding corpus-based concatenative synthesis for Mokusei domain • Refine system component capabilities: – More robust acoustic and language models for ASR – Improved coverage for language understanding • Expansion of weather content for Japan NTT - MIT Research Collaboration — Bi-Annual Report, January 1—June 30, 2000

Recommend


More recommend