International Journal of Scientific Research and Engineering Development

The rapid growth of artificial intelligence has made it possible to generate highly realistic synthetic voices that closely resemble human speech. While this technology has many useful applications, it has also increased the risk of voice-based fraud, identity theft, misinformation, and unauthorized access. These challenges highlight the need for accurate and efficient deepfake voice detection systems. This paper presents VocaTrust: Machine Learning-Based Voice Integrity Assessment, a framework designed to detect manipulated speech and evaluate voice authenticity. The system uses genuine audio samples from the LibriSpeech dataset and spoofed audio from the ASVspoof dataset. After preprocessing, Mel-Frequency Cepstral Coefficients (MFCCs) are extracted from each audio sample to capture key speech characteristics. These features are then used to train a Convolutional Neural Network (CNN) that classifies speech as genuine or fake. In addition, the model generates a Voice Integrity Score that indicates the likelihood of the audio being authentic.
The proposed framework is lightweight, computationally efficient, and suitable for real-time applications using only audio input. By providing reliable deepfake voice detection with low computational overhead, VocaTrust can support secure voice authentication, banking systems, digital forensics, and voice-enabled applications against emerging spoofing attacks.

Prof.Yashodara R, Deeksha, Dimple G, Sinchana Gowda RP, Varsha NG, "VocaTrust: Machine Learning Based Voice Integrity Assessment" International Journal of Scientific Research and Engineering Development, V9(4): Page(824-829) July-August 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.