International Journal of Image Processing and Pattern Recognition
AI-Based Deepfake Voice Detection Model for Hindi Language
1 Research Scholar, MCA Department Thakur Institute of Management Studies & Career Development & Research Mumbai, India
2 Research Scholar, MCA Department Thakur Institute of Management Studies & Career Development & Research Mumbai, India
3 Professor, MCA Department Thakur Institute of Management Studies & Career Development & Research Mumbai, India
Abstract
Most current research on deepfake voice detection focuses heavily on English or Mandarin, creating a significant security gap for Hindi speakers. To address this, we developed a detection framework specifically tailored for the Hindi language using a lightweight approach: Mel-Frequency Cepstral Coefficient (MFCC) feature extraction paired with an XGBoost classifier. We trained the system on a large dataset comprising over 99,000 real voice samples and roughly 3,800 synthetic samples, each clipped to a duration of 5–8 seconds. The model proved highly effective against standard text-to-speech (TTS) audio, achieving 99.99% accuracy with perfect precision and recall scores on our test set. We further stress- tested the system by introducing background noise; even at a high noise level (5dB SNR), the model maintained a robust 88% accuracy. However, we discovered a key limitation: while the model easily flags standard TTS deepfakes, it struggles to distinguish hyper- realistic clones generated by advanced tools like ElevenLabs. This study establishes a crucial baseline for Hindi audio forensics while highlighting that future work must evolve beyond classical features to counter high-fidelity AI generation.
Keywords— Deepfake Voice Detection, Hindi Speech, Artificial Intelligence, MFCC, XGBoost, Audio Forensics, Machine
Learning
References (25)
- Bird J, Lotfi A. Lightweight MFCC-based real-time detection of voice conversion. In: Proceedings of INTERSPEECH; 2020.
- Yamagishi J, et al. ASVspoof 2021 challenge: Evaluating countermeasures for synthetic speech. In: Proceedings of the IEEE Spoken Language Technology Workshop; 2021.
- Chen H, et al. DeepSonar: Deep neural network-based speech deepfake detection via neuron behavior analysis. In: Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP); 2021.
- Mittal A, et al. Audio deepfake detection using MFCC and CNN. IEEE Access. 2022;10:23411-23420.
- Lim S, et al. Explainable AI for audio deepfake detection. IEEE Access. 2022.
- Pianese G, et al. Deepfake audio detection by speaker verification. arXiv [Preprint]. 2022. Available from: https://arxiv.org/abs/2209.14098
- Wu Z, et al. Spoofing detection in automatic speaker verification: CQCC and GMM approach. IEEE/ACM Trans Audio Speech Lang Process. 2023.
- Yi H, et al. Audio deepfake detection: A survey. arXiv [Preprint]. 2023. Available from: https://arxiv.org/abs/2308.14970
- Zhang X, et al. What to remember: Self-adaptive continual learning for audio deepfake detection. arXiv [Preprint]. 2023. Available from: https://arxiv.org/abs/2312.09651
- Liu X, et al. Speaker-aware anti-spoofing. In: Proceedings of INTERSPEECH; 2023.
- Guo Y, et al. Audio deepfake detection with self-supervised WavLM and multi-fusion attentive classifier. arXiv [Preprint]. 2023. Available from: https://arxiv.org/abs/2312.08089
- Zhang Y, et al. D-CAPTCHA: Interactive audio challenge for deepfake detection. IEEE Trans Inf Forensics Secur. 2023.
- Pitch M. Human-AI collaboration for scalable deepfake detection. arXiv [Preprint]. 2024. Available from: https://arxiv.org/abs/2402.11234
- Xie Z, et al. FakeSound: Deepfake general audio detection. In: Proceedings of INTERSPEECH; 2024.
- Li X, et al. SafeEar: Content privacy-preserving audio deepfake detection. arXiv [Preprint]. 2024. Available from: https://arxiv.org/abs/2409.09272
- Anonymous. A robust unified spoofing audio detection scheme. Comput Electr Eng. 2024.
- Anonymous. Audio deepfake detection using spectrogram-based feature and CNN model. arXiv [Preprint]. 2024. Available from: https://arxiv.org/abs/2407.01777
- Anonymous. Audio deepfake detection: What has been achieved and what lies ahead. Sensors. 2024.
- Anonymous. A survey on speech deepfake detection. arXiv [Preprint]. 2024. Available from: https://arxiv.org/abs/2404.13914
- Anonymous. Investigation of deepfake voice detection using speech pause analysis. JMIR Biomed Eng. 2024.
- El Kheir Y, et al. Comprehensive layer-wise analysis of SSL models for audio deepfake detection. In: Findings of the Association for Computational Linguistics: NAACL; 2025.
- Anonymous. Audio deepfake detection using MFCC features. Int J Future Multidiscip Res. 2025.
- Anonymous. Towards reliable audio deepfake attribution and model recognition: A multi-level autoencoder-based framework. arXiv [Preprint]. 2025. Available from: https://arxiv.org/abs/2508.02521
- OpenSLR. Hindi speech dataset [Internet]. Available from: https://www.openslr.org/103/
- IISc Speech Processing Lab. Hindi male and female TTS datasets [Internet]. Available from: https://syspin.iisc.ac.in/datasets