Evaluation of speech quality measures for the purpose of speaker verification
Jonas Richiardi and Andrzej Drygajlo
Abstract
Real-world deployment of speaker verification systems often have to contend with degraded signal quality and erratic statistical behaviour of the speech data being modelled. We present signal quality estimation techniques for extraction of additional information about the speech data that can be used to improve performance of speaker verification systems in degraded conditions. We propose methods to perform objective evaluation of these quality measures for the purpose of their comparison using benchmarking databases, and show why the class must be taken into account when evaluating quality measures.
|
|
|||||
|
|
|
|
|
|
