EUROSPEECH 2003 - INTERSPEECH 2003 8th European Conference on Speech Communication and Technology Geneva, Switzerland September 1-4, 2003

Tfarsdat - The Telephone Farsi Speech Database

Mahmood Bijankhan (1), Javad Sheykhzadegan (2), Mahmood R. Roohani (2), Rahman Zarrintare (2), Seyyed Z. Ghasemi (1), Mohammad E. Ghasedi (2)

(1) University of Tehran, Iran
(2) Research Center of Intelligent Signal Processing, Iran

This paper describes an ongoing research to create an acoustic phonetic based telephone Farsi speech database, called "Tfarsdat". It is compared with two LDC Farsi corpora, OGI and Call friend in terms of corpus dialectology. Up to now, we have recorded about 8 hours of monologue calls containing spontaneous and read speech for 64 speakers belonging to one of ten dialect regions. A hierarchical annotation system is used to transcribe phoneme, word and sentence levels of speech data. User software is written to access speech and label files efficiently using a menu driven query system. We conducted two experiments to validate Tfarsdat statistically. Results showed the necessity of increasing speaker size and also quality enhancement of annotation system.

Full Paper

Bibliographic reference.  Bijankhan, Mahmood / Sheykhzadegan, Javad / Roohani, Mahmood R. / Zarrintare, Rahman / Ghasemi, Seyyed Z. / Ghasedi, Mohammad E. (2003): "Tfarsdat - the telephone farsi speech database", In EUROSPEECH-2003, 1525-1528.