This project focused on the harder version of audio deepfake detection: identifying which parts of an utterance are forged, not only whether the whole clip is synthetic. The work adapted speech deep learning pipelines for PartialSpoof-style data, studied the gap between utterance-level detection and frame-level localization, and turned the results into a research-style report and poster.
Speech AI / 2025
Identification of Authentic and Forged Segments in Partially Spoofed Speech
Adapted deep learning architectures to detect synthetic regions inside partially spoofed speech audio, combining speech feature extraction, spoof-detection modeling, evaluation, and research writing around the harder problem of localizing forged segments.