View article

Speaker verification based on phonetic-acoustic approach and text-dependent framework has been applied for forensic purposes in Indonesian court since 2008. In order to accelerate the speaker verification process, an automatic text-independent system is developed. This automatic system employs MFCC features and GMM speaker modeling, a standard and simple approach used in automatic speaker recognition. The system is developed and evaluated in a speech database spoken in Bahasa Indonesia. The database is recorded in condition that mimic the typical real forensic speaker recognition cases in Indonesia. In this first development stage, the automatic speaker verification system has successfully achieved the EER of 6.45% for male speakers and 12.9% for female speakers.