Advances in AI-based voice production and conversion technologies have made it possible to create deepfake voices that closely resemble real human speech, raising new security challenges in forensic voice analysis and cybersecurity. Traditional forensic audio analysis methods rely primarily on acoustic characteristics, which can be limited by the increasing realism of deepfake audio. This study proposes an approach for forensic audio analysis that considers the temporal structure of speech, prosodic features and inconsistencies between audio and textual content to enhance the detection of deepfake audio. Accordingly, a dataset was created containing a total of 60 audio recordings, consisting of 30 real and 30 artificial Turkish voice recordings, each 7 - 10 seconds long. Mel-Frequency Cepstral Coefficients (MFCCs) were extracted from each audio recording; text and word time tags were obtained using an automated speech recognition method. Based on these time tamps, speech rate, pause durations and temporal alignment features were calculated. In addition, prosodic features such as pitch and amplitude were incorporated into the model. All obtained features were classified in two stages using the Random Forest classification algorithm. The model was analyzed in two stages: firstly, without prosodic features, and secondly, with the addition of prosodic features. Model performance was then evaluated using 5-fold cross-validation. The results indicate that incorporating prosodic features leads to higher deepfake Detection accuracy compared to models that rely solely on non-prosodic features. These results aim to demonstrate that temporal and prosodic inconsistencies in forensic audio analysis provide supportive and complementary elements for deepfake audio detection.
References
[1]
Chesney, R. and Citron, D. (2019) Deepfakes and the New Disinformation War. Foreign Affairs, 98, 147-155.
Kıvanç, T. and Çötok Akıncı, N. (2024) Dijital iletişim ve deepfake. In: İşman, A., Öztunç, M. and Akıncı, Çötok N., Eds., Iletişim Çalişmalari. Eğitim Yayınevi, 66-74.
[4]
Sever, H. (2008) Adli ses incelemeleri ve hukuki boyutu. Adalet.
[5]
Ullah, R., Asghar, I., Malik, H., Evans, G., Ahmad, J. and Roberts, D.A. (2024) Enhancing Forensic Audio Transcription with Neural Network-Based Speaker Diarization and Gender Classification. 2024 International Conference on Engineering and Emerging Technologies (ICEET), Dubai, 27-28 December 2024, 1-6. https://doi.org/10.1109/iceet65156.2024.10913726
[6]
Liu, L., Li, W., Morris, S. and Zhuang, M. (2023) Knowledge-Based Features for Speech Analysis and Classification: Pronunciation Diagnoses. Electronics, 12, Article 2055. https://doi.org/10.3390/electronics12092055
[7]
Muda, L., Begam, M. and Elamvazuthi, I. (2010) Voice Recognition Algorithms Using Mel Frequency Cepstral Coefficient (MFCC) and Dynamic Time Warping (DTW) Techniques. arXiv:1003.4083.
[8]
Rabiner, L. and Schafer, R. (2010) Theory and Applications of Digital Speech Processing. Prentice Hall Press.
[9]
Yan, Y., Simons, S.O., van Bemmel, L., Reinders, L.G., Franssen, F.M.E. and Urovi, V. (2025) Optimizing MFCC Parameters for the Automatic Detection of Respiratory Diseases. Applied Acoustics, 228, Article 110299. https://doi.org/10.1016/j.apacoust.2024.110299
[10]
Wagner, M. and Watson, D.G. (2010) Experimental and Theoretical Advances in Prosody: A Review. Language and Cognitive Processes, 25, 905-945.
[11]
Jurafsky, D. and Martin, J.H. (2025) Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition with Language Models. 3rd Edition, Prentice Hall PTR, 4-37. https://web.stanford.edu/~jurafsky/slp3
[12]
Biau, G., Deevroye, L. and Lugosi, G. (2008) Consistency of Random Forests and Other Averaging Classifiers. Journal of Machine Learning Research, 9, 2015-2033.
[13]
Shanmugam, M., Ismail, N.N.N., Magalingam, P., Hashim, N.N.W.N. and Singh, D. (2023) Understanding the Use of Acoustic Measurement and Mel Frequency Cepstral Coefficient (MFCC) Features for the Classification of Depression Speech. In: Al-Sharafi, M.A., Al-Emran, M., Tan, G.WH. and Ooi, KB., Eds., StudiesinComputationalIntelligence, Springer, 345-359. https://doi.org/10.1007/978-3-031-48397-4_17
[14]
Hamza, A., Javed, A.R.R., Iqbal, F., Kryvinska, N., Almadhor, A.S., Jalil, Z., etal. (2022) Deepfake Audio Detection via MFCC Features Using Machine Learning. IEEEAccess, 10, 134018-134028. https://doi.org/10.1109/access.2022.3231480
[15]
Bohara, R. and Bairwa, A.K. (2025) Detecting Deepfake Audio Using Spectrogram-Based Machine Learning Approaches. IEEEAccess, 13, 149478-149489. https://doi.org/10.1109/access.2025.3602531
[16]
Iqbal, F., Abbasi, A., Javed, A.R., Jalil, Z. and Al-Karaki, J. (2022) Deepfake Audio Detection via Feature Engineering and Machine Learning. In: Drakopoulos, G. and Kafeza, E., Eds., Proceedings of the CIKM 2022 Workshops. CEUR Workshop Proceedings, Vol. 3318. https://ceur-ws.org/Vol-3318/paper4.pdf
[17]
Ali Sharafat, H.M., Rivzi Muslim, S.M., Tariq, H., Majeed, S., Tariq, A. and Iq-bal, M.M. (2025) AI-Based Deepfake Audio Detection Technique from Real and Fake Audio Dataset. JournalofComputing&BiomedicalInformatics, 8.
[18]
C, S.K., G, S.N.R., Patnam, V., Deepak, S., M. V, S.P. and G, T.R. (2024) Leveraging Deep Learning Architectures for Deepfake Audio Analysis. Proceedings of 6th International Conference on Smart Systems and Inventive Technology, 19, 266-276. https://doi.org/10.29007/sl8m
[19]
Todisco, M., Wang, X., Vestman, V., Sahidullah, M., Delgado, H., Nautsch, A., etal. (2019) ASVspoof 2019: Future Horizons in Spoofed and Fake Audio Detection. Interspeech 2019, Graz, 15-19 September 2019, 1008-1012. https://doi.org/10.21437/interspeech.2019-2249
[20]
Gessinger, I., Raveh, E., Steiner, I. and Möbius, B. (2021) Phonetic Accommodation to Natural and Synthetic Voices: Behavior of Groups and Individuals in Speech Shadowing. SpeechCommunication, 127, 43-63. https://doi.org/10.1016/j.specom.2020.12.004
