METU Turkish Microphone Speech Corpus

In this paper, work on developing a Turkish microphone speech corpus at the Middle East Technical University (METU) is presented. Before collecting the audio corpus, sound properties of Turkish have been investigated and a triphone-balanced set of Turkish sentences have been developed. Speech from 193 speakers, each uttering 40 sentences selected from the balanced sentence set, has been collected. The corpus has been aligned by the Turkish phoneme aligner developed. Each speech file is associated with phoneme, HMM state and word level alignments. In addition to these, each speaker has a text file containing the age, region, gender, education and etc. information and also the uttered sentences. The aim of collecting such a corpus is to obtain a standard and common microphone speech corpus for Turkish speech research. The corpus is open for research purposes. It has also been accepted to be distributed by the Linguistic Data Consortium of the Pennsylvania University in November 2005.