Influence of GSM coding on speaker recognition using short polish sequences

摘要

The paper presents a comparison of speaker models used for fast speaker identification in short recordings of telephone conversations. The knowledge of the encoder type used during the transmission of speech allows to apply a model that takes specific characteristics of the encoder into account. This improves efficiency of the speaker recognition process. The influence of the following GSM encoders was tested: FR, HR, EFR and AMR. During the experimental research we used our database of short voice phrases that usually occur during emergency calls. This paper is based on our studies related to techniques for the GSM encoding detection and to algorithms for removing silence in the voice recordings.