SUMMEETING: AUTOMATIC MEETING TRANSCRIPTION AND SUMMARIZATION WITH AN ADAPTED ONTOLOGY FOR THE LEGAL DOMAIN AND BUSINESS ADMINISTRATION

Every organization use internal meetings as a key instrument and communication channel among their workers. Meetings are also critical in inter-organizational communication. Currently, European companies can choose between two options depending on the degree of detail that they need: To devote their own resources and effort by having a worker taking notes during the meeting and preparing a minutes after the meeting. To contract a costly Transcription service. SUMMEETING will provide an affordable tool for SMEs that will allow them to get their own automatic meeting transcriptions and will allow transcription services providers to offer reviewed transcriptions at a lower price.

Meeting transcription has been a relevant field of research during a long time and results have been improving as the speech technologies had improve. However, in the main transcription evaluation event, the Rich Transcription Evaluation conference, it has been proven that the hypothesis that suggest that traditional speech processing techniques can not help to improve current levels of success in meeting transcription was right as the results of the evaluation in 2007 and 2009 (this evaluation conference takes place once every 2 years) show no significant progress. This hypothesis suggests that for improving the quality of automatic transcription of meetings it is necessary to use a priori information about the meeting content. The objective of this project is to focus in the domain of legal meeting in order to reduce the vocabulary and take advantage of the structure information present in meetings. In this way, the SUMMEETING project will be able to improve the current results of meeting transcription from a current Word Error Rate: 20% to a WER of 10%. In addition, the project will contribute to the field of automatic summarization of texts by focusing on the field of legal meeting summarization and by studying the effect of transcription errors in the later automatic summarizations. It is important to highlight that a priori information related with structure and special vocabulary used in the transcription will be also of key importance in the automatic summarization process. Additionally the project will develop also a hardware system that will allow the use of wireless communication (Zigbee) between the different headsets and the PC recording the audio of the meeting by implementing a Distributed Speech Recognition codec in each of the headsets.

Automatic Speech Recognition
Automatic Summarization
USB
microphones
audio recorder

RTD (coordinator) - Spanish R+D center RTD - European University expert in Automatic Speech Recognition 2 EUROPEAN SMEs WITH A TECHINICAL ROLE