Speech Processing Group česká verze
english version
Welcome | History | Staff | Research | Projects | Publications | Download | Demos | Relations | Links | Contact Us

History of the group

The speech processing group was established in 1981. From the very beginning the research took place in the speech synthesis and analysis. The wave shape coding algorithms were a vital topic and the results of this research found application in several synthesizers with a limited vocabulary and high quality of synthetic speech. Then the group was very active in the filed of speech recognition in several following years. Hidden Markov model based isolated digits speaker independent recognizer implemented on THS320E17 was the latest achievement from this period. Later the research was extended to further problems in the field of speech recognition, e.g. the improvement and design new types of descriptors for speech recognition, the enlargment of the number of recognized words going to the current large vocabulary countinuous speech recognition systems. Also the applications of neural networks within speech recognition systems were studied, mainly at the level of feature extraction. Continuously, great emphasis is given to linguistic and phonetic background, which is important part of recognition research.

As a continuation of the work on general speech analysis, the behavior of designed systems working with the corrupted speech or speech with noisy background has started being studied since 1992. The first solved project focused on speech enhancement for communication purposes was realized in the co-opperation with US WEST company. Within these research activities, also the speech pre-processing for hearing aids and cochlear implants was studied. The research in the fields of speech recognition and speech enhancement yielded logically into the research in the field of robust speech recognition, i.e. the recognition of speech under real conditions, which became one the most important part of our current activities. Special attention is devoted to car environment and telephony speech recognition.

Since 1999 the activities are focued also to collection of large speech database from real environment. The group was involved actively into several European projects, namely INCO-Copernicus 977017 Eastern European Speech Databases for Creation of Voice Driven Teleservices (SpeechDat-E) and FP4 IST-1999-10003 SPEECON - Speech Driven Interfaces for Consumer Applications and as the results of the participation within these projects, Czech versions of databases SpeechDat a SPEECON were created. These databases are available via ELRA (see Catalogue). Further databases were created within bilateral projects with European industrial companies or academic labs (e.g. Siemens AG, ETMIC, Harman/Becker, SVOX, Radboud University in Nijmegen, etc.).

Our research is standradly supported by grants, mainly within Czech Grant Foundation Agency projects solved since 1996, which contributed to a significant progress in the field of speech recognition also on the international level. It was project: GACR 102/96/K087 Theory and Application of Speech Communication in Czech (1996-2001), GACR 102/02/0124 Voice Technologies Supporting Information Society (2002-2004), GACR 102/05/0278 New Trends in the Research and Application of Speech Technologies (2005-2007), GACR 102/08/0707 Speech Recognition under Real-Word Conditions (2008-2011). The group participated also within European research project of COST family, mainly in COST 249: Continuous Speech Recognition over the Telephone and in the following COST 278: Spoken Language Interaction in Telecommunication.

Last updated Tue Apr 15 17:35:37 CEST 2014 Mail any suggestion or problem to webmaster. Maintenance.