تأیید بلندگو از سخنرانی تحریف شده در کدک برای تحقیقات قانونی با ترکیب سریال طبقهبندیها

Forensic investigation often uses biometric evidence as important aids for identifying the culprits. Speech is one of the easily available biometrics in today's hi-tech world. But, most of the speech biometric evidence acquired for investigative purposes will usually be highly distorted. Among these distortions, most prominent is the distortion introduced by the speech codec. Speech codec may either remove or distort some of the speaker-specific features, and this may reduce the speaker verification accuracy. The effect of distortion on commonly used speaker-specific features namely Mel Frequency Cepstral Coefficients (MFCC) and Power Normalized Cepstral Coefficients (PNCC), due to Code Excited Linear Prediction (CELP) codec (the most widely used speech codec in today's mobile telephony), is quantified in this paper. The features which are least affected by the codec are experimentally determined as PNCC. But, when these PNCC coefficients are directly employed, speaker verification error rate obtained is 20% with Gaussian Mixture Model-Universal Background Model (GMM-UBM) classifier. To improve the verification accuracy, PNCCs are slightly modified, and these modified PNCCs (MPNCC) are used as the feature set for the speaker verification. With these modified PNCCs, the error rate is reduced to 15%. By fusing these MPNCCs with MFCC, the error rate is further reduced to 8.75%. A series combination of GMM-UBM and Support Vector Machine (SVM) classifiers is also proposed here to enhance the speaker verification accuracy further. The speaker verification error rates for different baseline classifiers are compared with that of the proposed serially combined GMM-UBM and SVM classifiers. The classifier fusion with the fused feature set largely reduced the error rates to 2.5% which is very much less than that of baseline classifiers with normal PNCC features. Hence, this system is a good candidate for investigative purposes.