Serpens for Kepler
Advanced Automatic Speech-to-Text Conversion System Dedicated for Internal Security Agencies
The goal of the project is to design and implement advanced speech-to-text conversion system dedicated to the agencies responsible for the homeland security. The system will offer the functionality of converting speech to text for Polish language, independent of the speaker and the vocabulary used by that speaker. Conversion to text will also be performed for speech recorded prior to processing under various acoustic conditions. Moreover, the system will perform large recording database indexing in order to enable efficient word phrase search.
The system operation will be guided by the needs of the end users, namely the internal security agencies, as far as the functionality, vocabulary and the utterance context are concerned. Such an approach will improve speech-to-text conversion accuracy.
Scheduled finish date: 2016-09-30
Conversion to text will be performed for utterances dictated in the office-like environment as well as recorded prior to processing in various locations and under various conditions, in order to expedite the process of preparing memos, reports of inspections and other activities conducted by the security agencies officials. In addition to dictation also spontaneous speech will be processed and converted to text. This type of utterances is important for transcribing security agencies briefings and meetings, as well as reporting field activities. Consequently, conversion to text will be performed on the recordings obtained under favorable acoustic conditions which do not contain significant background noise, as well as for recordings obtained under more difficult conditions characteristic for field-like environment.
The system will be compatible with the most commonly used tools such as Microsoft Office, and will be integrated with applications and services typically used by the security agencies.
The speech-to-text conversion system will offer the following functionality:
- conversion of dictation,
- conversion of speech recorded prior to conversion,
- conversion of recordings performed under various acoustic conditions,
- automatic large recording database indexing for efficient phrase search,
- recognition of the selected linguistic and non-linguistic features for speaker profile and circumstances analysis.
The formation of mechanisms of EU projects related to the ICT industry was a leading topic of the "European Consortia 2011-2012 - development of clustering in Wielkopolska" Conference which was held in Poznań. The conference was organized jointly by the Marshal Office of the Wielkopolska Region, Poznan Supercomputing and Networking Center and Wielkopolska ICT Cluster.