Catalogue on Research and Innovation
Potential of University in Novi Sad

The main objective of the “AlfaNum” team is to reaffirm the leading position in the development of quality speech technologies for most South Slavic languages and their applications in the Western Balkans. The main activities of “AlfaNum” are:
• The development of high-quality and more advanced text-to-speech synthesis (TTS)
• The development of large vocabulary continuous automatic speech recognition (ASR)
• The development of speaker identification and verification systems
• The research and development of emotion speech recognition
• The development of natural language processing modules including dialogue management
• The application of the developed speech technologies in Western Balkan countries:

• in multimodal human-machine dialogue systems (IVR, smart phones, smart homes)
• for purposes such as: text reading, text dictation, speech transcription
• within aids for the physically disabled, visually impaired, speech impaired, hearing impaired

• “AlfaNum” has already developed both small-to-medium vocabulary ASR and high-quality concatenationbased TTS in Serbian, Croatian and Macedonian.
• A number of valuable speech and language resources for Serbian and kindred South Slavic languages have been created within several projects over the last decade. Apart from these resources, a number of expert systems, machine learning systems as well as mathematical models have been developed and deployed in the first speech enabled products in Serbia, Croatia, Bosnia and Herzegovina, Montenegro, and Macedonia - the countries where kindred South Slavic languages are predominantly spoken. For example, one can listen to news at a number of speech-enabled web sites (Radio Television of Serbia - RTS, Radio Television of Vojvodina - RTV, eUprava, as well as several municipalities) using a computer or a smart phone. The visually impaired can listen to any text displayed on the screen using anReader (AlfaNumTTS). The AlfaNumASR and AlfaNumTTS components have provided smart phones with basic speech generation and understanding functionalities in Serbian.
• Further development of both large vocabulary ASR and more advanced TTS is based on the aforementioned speech and language resources. Both technologies will enable a much wider range of applications and will contribute to the preservation of Serbian and kindred languages in this new domain of human-machine speech communication.

• Delić V, Sečujski M, Jakovljević N, Janev M, Obradović R, Pekar D. (2010) “Speech Technologies for Serbian and Kindred South Slavic Languages”, In Advances in Speech Recognition, N. Shabtai (Ed.), SCIYO, pp. 141-164.
• Popović B, Janev M, Pekar D, Jakovljević N, Gnjatović M, Sečujski M, Delić V. (2012) “A Novel Split-and-Merge Algorithm for Hierarchical Clustering of Gaussian Mixture Models”, Applied Intelligence, ISSN: 0924-669X, Springer-Verlag New York, Inc., Vol. 37, No. 3, pp. 377-389, DOI: 10.1007/s10489-011-0333-9
• Gnjatović M, Janev M, Delić V. (2012) “Focus Tree: Modeling Attentional Information in Task-Oriented Human-Machine Interaction”, Applied Intelligence, ISSN: 0924-669X, Springer-Verlag New York, Inc., Vol. 37, No. 3, pp. 305-320, DOI: 10.1007/s10489-011-0329-5
• Delić V, Bojanić M, Gnjatović M, Sečujski M, Jovičić S. (2012) “Discrimination Capability of Prosodic and Spectral Features for Emotional Speech Recognition”, Electronics and Electrical
Engineering, ISSN 1392-1215, Vol. 18, No. 9, pp. 51-54, DOI:10.5755/j01. eee.18.9.2806.
• Jakovljević N, Mišković D, Janev M, Sečujski M, Delić V. (2013) “Comparison of Linear Discriminant Analysis Approaches in Automatic Speech Recognition”, Electronics and Electrical Engineering, ISSN 1392-1215 (in press).