BG-SRDat: A Corpus in Bulgarian Language for Speaker Recognition over Telephone Channels

Atanas Ouzounov
Institute of Information Technologies-BAS
Acad. G. Bonchev Str. bl. 29A,
Sofia 1113, Bulgaria,
E-mail: atanas@iinf.bas.bg

Abstract

In the paper is described the BG-SRDat (BulGarian language Speaker Recognition DATa) - a corpus in Bulgarian language collected over noisy analog telephone channels and intended for speaker recognition. The BG-SRDat comprises two different speech data, called Speech Data 1 (SD1) and Speech Data 2 (SD2), respectively. The SD1 is a reading text from a newspaper and its average length is about 40 seconds. The SD2 is a short phrase with length of about 2 seconds. The SD1 and the SD2 are uttered in various sessions by different number of speakers (male) – 26 and 13, respectively. To achieve more realistic real-world conditions the speech data is collected by different types of telephone calls (internal-routing, local and long-distance) and various acoustical environments (noisy offices, halls and streets). The main purpose of the BG-SRDat is to provide data for evaluation of various speaker recognition techniques with noisy telephone speech in Bulgarian language.