Confidence Measures for Speech/Speaker Recognition

When it comes to the speech recognition, confidence becomes a crucial word as speech recognition results are usually erroneous when you want to use a computer to transcribe a continuous speech. Measuring the confidence on speech recognition results is the main problem investigated throughout this book. The results and the methodology discussed in this study continue to be an important research topic for the scientific community. The study also contain some experiments and speech recognition results for Turkish language which is the native language of the author. Based on the morphological structure of the language a new language modeling technique has been proposed and analysed for Turkish. Agglutinative languages like Turkish, Korean, and Japanese would certainly benefit from the proposed language model for better speech recognition accuracies.

Autorentext

Erhan Mengusoglu holds a B.S. in Computer Science and Engineering (1995) and a M.Sc. in Computer Science (1999) from Hacettepe University, Turkey. He obtained his PhD at Faculte Polytechnique de Mons (Belgium) in May 2004. His thesis work consisted in exploring different kind of confidence measures speech and speaker recognition.