Transcript PowerPoint bemutató
Applied Speech Information Systems
2/008 NRDP Project
Project coordinator:
Prof. Géza Gordos ([email protected])
Members of the consortium:
• Budapest University of Technology and Economics, Department of Telecommunications and Telematics (BUTE) • Pázmány P. Catholic University (PPCU) • Westel Mobile Telecommunications Company (WMT) • AITIA Inc. (AITIA) • MorphoLogic Ltd. (ML) 4/25/2020 Applied Speech Information Systems (Contact: [email protected]) 1/14
General objectives
• Integration of generic infocommunication technologies • Extension of complementary knowledge bases – 2 university research labs – 2 ICT SME companies – 1 telecommunication corporation • Instead of expensive and often poor quality adaptation of technologies developed for the English language, the characteristics of the Hungarian language are taken into account • Strengthening international competitive position • Assistance in integration of handicapped people 4/25/2020 Applied Speech Information Systems (Contact: [email protected]) 2/14
Technical objectives
•
The innovative goal of the project:
research, development and demonstration of generic technologies integrated into practical services •
New generic technologies:
speaker independent, open dictionary Hungarian speech recognizer for telephone channel, based on speech databases and speech analysis (text based dictionary generation) speech synthesizer with variable size acoustic database units and general Hungarian name and address reader intelligent dialogue system description and management framework system applying speech interfaces integrating the above technologies for intelligent telecommunication information retrieval systems 4/25/2020 Applied Speech Information Systems (Contact: [email protected]) 3/14
Technical objectives and results
Speech recognition (BUTE – AITIA)
• Goal: Development of multi-channel speaker independent speech recognition engine for telephone services
Completed subtasks
– speech database generation (600 speakers x 5 minutes, noisy) – development of an automatic segmentation method – optimization of the reliability and resource requirements of the recognition algorithm (PC: 20 channels, DSP: 180 channels front-end) – training of phone models (monophone, diphone, triphone) – automatic phonetic transcription for dictionary generation 4/25/2020 Applied Speech Information Systems (Contact: [email protected]) 4/14
Technical objectives and results
Speech synthesis (BUTE)
• Goal: Development of – improved high quality text-to-speech framework – generic name and address reader for Hungarian – application(s) to special database access
Completed subtasks
– processing of 3+2 million name and address records • manual classification of 300.000 records – automatic classifier of Hungarian proper/company names and addresses – new name and address reading dialogue strategies – detailed reading (syllabification and spelling) algorithms – new TTS + name and address reading system 4/25/2020 Applied Speech Information Systems (Contact: [email protected]) 5/14
Technical objectives and results
Dialogue system (AITIA)
• Goal: Development of an intelligent framework and dialogue system for speech recognition based call center and voice portal applications
Completed subtasks:
– speech recognition based dialogue management system – development of the dialogue description structure – implementation of the dialogue editor – implementation of the dialogue manager 4/25/2020 Applied Speech Information Systems (Contact: [email protected]) 6/14
Research Objectives and Results
Individual voice features (PPCU)
• Goal: Analysis and modification of individual speech feature characteristics
Completed subtasks
– analysis of speaker voice timbre and features – transplantation of source speaker's features into target speaker's voice 4/25/2020 Applied Speech Information Systems (Contact: [email protected]) 7/14
Research Objectives and Results
Language model (ML)
• Goal: Study and report on integration of linguistic language models into speech controlled applications
Completed subtasks
– analysis of application of possibilities for feedback of linguistic analysis in speech recognition – analysis of possibilities of linguistic support for increasing the recognition rate of the most probable character string 4/25/2020 Applied Speech Information Systems (Contact: [email protected]) 8/14
Results of integration Voxenter
TM
voice portal
• Voxenter is connecting to and extending the functionality of any PBX or call center • Flexible agent-based real-time information service management and database connectivity • Unique and standards-based dialogue editor supporting XML and VXML • Web/Java based remote administration console • It can be connected to analogue, ISDN (BRI, PRI), and VoIP telephone interfaces • AITIA Inc. and DTT-BUTE use the system since December 2002 (call +36 1 382-7580) • IT 2003 Hungary conference and exhibition award 4/25/2020 Applied Speech Information Systems (Contact: [email protected]) 9/14
speech databases
Voice portal development
training phone models language models synthesis units synthesis rules speech recognition engine speech synthesizer dialogue system telephone interface speech characteristics analysis and modification at development: 4/25/2020 dialogue editor operator framework system dialogue manager dialogue description Applied Speech Information Systems (Contact: [email protected]) database 10/14
Results of integration Speech enabled call center
Integration of speech recognition and speech synthesis systems with AVAYA technology and development of demo applications for Westel Inc.
– Billing information service (integrating speech recognition and number and date reader) – Telephone number based reverse directory assistance (integrating name and address reader) 4/25/2020 Applied Speech Information Systems (Contact: [email protected]) 11/14
Avaya call center integration
phone models language models NLSR protocol interface speech recognition engine AVAYA call center 4/25/2020 Proxy TTS protocol interface ProfiVox speech synthesizer synthesis units synthesis rules Applied Speech Information Systems (Contact: [email protected]) 12/14
Project exploitation
Research, education
– new scientific results, strengthening of graduate and PhD schools – growing international collaboration potential – feedback into the education
Industry
– opening new market possibilities – integrating new languages (
industrial demand for other central-european languages TTS, ASR
) – further products and integrations (
100 free ProfiVox TTS licences for blind people, Digitania, T-Systems RIC, ...
) 4/25/2020 Applied Speech Information Systems (Contact: [email protected]) 13/14
Thanks for your attention!
4/25/2020 Applied Speech Information Systems (Contact: [email protected]) 14/14