The growing sophistication of speech generated by Artificial Intelligence (AI) has introduced new challenges in audio deepfake detection. Text-to-speech (TTS) and voice conversion (VC) technologies can now produce convincing synthetic speech with high quality and intelligibility. This poses a serious threat to voice biometric security systems, such as automatic speaker recognition. It also increases the risks associated to the spread of spoken disinformation, where synthetic voices can be used to disseminate malicious content. In this study, we conduct an analysis of real and synthetic speech at phonetic and word levels. For that, a parallel dataset comprising real and synthetic speech signals were developed based on a subset of the LibriSpeech ASR corpus. Synthetic speech samples were generated using two TTS and one VC systems: Coqui TTS, VITS TTS, and StarGANv2 VC. We adopted HuBERT, a self-supervised speech model, to extract speech embeddings. The motivation for using this model stems from its ability to recognize sound units corresponding to the so-called pseudo phonemes. Our analysis is based on the KL divergence (KLD) between the distributions of synthetic and real phonemes, which allowed us to rank synthetic phonemes based on their alignment with their real counterpart. We also trained several classifiers per phoneme to distinguish between real and synthetic samples. We then compute the correlations between KLD and accuracies per phoneme. Besides showing a list of phonemes that are more discriminative, our findings suggest that vowels correlate better with the classifiers’ performance, suggesting that the KLD can be an indicator of the most distinguishable phonemes for deepfake detection.

Usual handwritten signature verification systems address the writer-independent (WI) approach using only bi-class robust classifiers to deal with the most challenging tasks. Indeed, WI concept, the reduced size of references and one-class signature verification are still open issues in practical cases. In this paper, we propose a one-class WI system using feature dissimilarity measures thresholding for classification and a reduced number of references. The proposed system involves the use of contourlet transform-based directional code co-occurrence matrix feature generation method. The verification is achieved through a WI threshold which is automatically selected using a new signature stability criterion. The proposed WI concept is besides addressed through the mixture of different writers' data sets in both the design and verification stages. Experimental results show the effectiveness of the proposed system in spite of the strict verification protocol using the one-class concept, a unique threshold for accepting or rejecting a questioned signature, the reduced number of writers, and the limited number of reference signatures.

We address, in this work, a new feature generation method for two different approaches of off-line handwritten signature verification (HSV), writer-dependent and writer-independent HSV. The proposed method uses conjointly the contourlet transform and the co-occurence matrix. The contourlet transform allows capturing contour segment directions of the handwritten signature, while the co-occurrence matrix allows describing the number of directions. Experiments are conducted on the well known CEDAR dataset and the classification through the support vector machines (SVM). The obtained results show the effective use of the Contourlet transform for handwritten signature verification comparatively to the state of the art.