Speaker Recognition Based On Breath Biometrics

Speaker Recognition is a generic term used for two problems:Speaker Identication and Speaker Verification.The speech contents are categorized as text-dependent,text-independent and text-prompted.Speaker Recognition is realized through Training/Enrollment phase which computes verified speaker reference model from feature vectors,and Testing/Verification phase which finds the similarity score between the testing speech from the unknown speaker and verified speaker reference.Over last two decades,Speaker Recognition has primarily been focused on source,system,and prosodic features of the speech.The breath,however,has either been treated as a trivial part of the speech,or considered a noise entity.Our observation reveals that breath is a unique fingerprint of human respiratory system which offers overwhelming results for Speaker Recognition.Moreover,its passive nature,short-duration,fewer occurrences and simple processing results to a light-weight,text-independent and transparent system,which we articulate as Breath ID.Breath ID scheme comprises of Breath Demarcation,Feature Extraction and FeatureMatching stages.According to the CDF analysis and empirical study,the breath features are extracted and classified by Mel Frequency Cepstral Coefficients,MFCC,based template matching technique.The verification is performed by a similarity based scheme,whose efficiency competes with a light-weight classification algorithm,which is based on a series of basic vector operations.Breath ID is evaluated by two steps in the thesis.Firstly,a data set collected from 50 users is processed.Our system offers a 0.04% False Identification Rate,FIR,for Speaker Identification,and 0.12% False Acceptance Rate,FAR,and 0.15% False Rejection Rate,FRR,for Speaker Verification.Secondly,the scheme further is evaluated under various practical modalities by another 20 users,like text in-dependence,replay scenario,users' motion status(sitting and walking),recording equipment(03 smartphones and 02 microphones),recording period(08 months),and bilingual contents(English and Chinese).