Koncepcja metody modyfikacji ukrytych reprezentacji głosu w celu jego rekonstrukcji na podstawie śladu pamięciowego świadka
Wybierz format
RIS BIB ENDNOTEData publikacji: 14.09.2026
Problems of Forensic Sciences (Z Zagadnień Nauk Sądowych), 2026, 145, s. 45-63
https://doi.org/10.4467/12307483PFS.26.003.23948Autorzy
Steering latent voice representations for voice reconstruction based on a witness’s memory trace
This article addresses the significant scarcity of forensic tools for generating voice samples for identification parades by proposing a novel method for voice reconstruction without a reference recording. The presented solution serves as an auditory analogue to facial composites, utilizing a hybrid architecture based on machine learning and speech synthesis (XTTS). The methodology combines an iterative algorithm for selecting a base voice candidate with a specialized neural network module that allows for the modification of interpretable acoustic parameters via latent space manipulation. Technical validation confirmed the system’s effectiveness in navigating the voice space and accurately translating physical parameters into vector representations. By shifting the identification burden from error-prone verbal descriptions to direct auditory perception, the system minimizes the verbal overshadowing effect. Consequently, the proposed prototype offers promising practical implications for forensic science, providing a technological foundation for law enforcement to conduct accurate voice lineups based solely on witness memory.
1. Juncu S, Fitzgerald RJ. A meta-analysis of lineup size effects on eyewitness identification. Psychol Public Policy Law. 2021;27(3):295–315. doi: 10.1037/law0000311
2. Steblay NM. Social influence in eyewitness recall: A meta-analytic review of lineup instruction effects. Law Hum Behav. 1997;21(3):283–297. doi: 10.1023/A:1024890732059
3. eLineup. Police photo lineup software [Internet]. eLineup; 2026. Available from: https://elineup.org/Default.aspx
4. Wells GL, Turtle JW. Eyewitness identification: The importance of lineup models. Psychol Bull. 1986;99(3):320–329. doi: 10.1037/0033-2909.99.3.320
5. Herbowski P, Słapczyńska D. Okazanie mowy w praktyce śledczej. Palestra. 2016;(7–8):43–50. doi: 10.19195/0137-1134.127.21
6. de Jong-Lendle G, Nolan F, McDougall K, Hudson T. Voice Lineups: A practical guide. Proc 17th ICPhS. 2015 Aug;1–5.
7. Pautz N, McDougall K, Mueller-Johnson K, Nolan F, Paver A, Smith HMJ. Identifying unfamiliar voices: Examining the system variables of sample duration and parade size. Q J Exp Psychol. 2023;76(12):2804–2822. doi: 10.1177/17470218231155738
8. Smith HMJ, Baguley T. Unfamiliar voice identification: Effect of post-event information on accuracy and voice ratings. J Eur Psychol Stud. 2014;5(1):59–68. doi: 10.5334/jeps.bs
9. Lavan N, Burston LFK, Garrido L. How many voices did you hear? Natural variability disrupts identity perception from unfamiliar voices. Br J Psychol. 2019;110:576–593. doi: 10.1111/bjop.12348
10. Read D, Craik FIM. Earwitness identification: Some influences on voice recognition. J Exp Psychol Appl. 1995;1(1):6–18. doi: 10.1037/1076-898X.1.1.6
11. Fenn KM, Shintel H, Atkins AS, Skipper JI, Bond VC, Nusbaum HC. When less is heard than meets the ear: Change deafness in a telephone conversation. Q J Exp Psychol. 2011;64(7):1442–1456. doi: 10.1080/17470218.2011.570353
12. McAllister HA, Dale RHI, Bregman NJ, McCabe A, Cotton CR. When eyewitnesses are also earwitnesses: Effects on visual and voice identifications. Basic Appl Soc Psychol. 1993;14(2):161–170. doi: 10.1207/s15324834basp1402_3
13. Belin P, Bestelmeyer PEG, Latinus M, Watson R. Understanding voice perception. Br J Psychol. 2011;102(4):711–725. doi: 10.1111/j.2044-8295.2011.02041.x
14. Bruce V, Young A. Understanding face recognition. Br J Psychol. 1986;77(3):305–327. doi: 10.1111/j.2044-8295.1986.tb02199.x
15. Frowd C, Hancock P, Carson D. EvoFIT: A holistic, evolutionary facial imaging technique for creating composites. ACM Trans Appl Percept. 2004 Jul;1(1):19–39. doi: 10.1145/1008722.1008725
16. George B, Gibson SJ, Maylin MI, Solomon CJ. EFIT-V – Interactive evolutionary strategy for the construction of photo-realistic facial composites. Proc 10th GECCO. 2008 Jul;1485–1490. doi: 10.1145/1389095.1389384
17. van den Oord A, Dieleman S, Zen H, Simonyan K, Vinyals O, Graves A, et al. WaveNet: A generative model for raw audio. arXiv [Preprint]. 2016 Sep. arXiv:1609.03499. doi: 10.48550/arXiv.1609.03499
18. Wang Y, Skerry-Ryan RJ, Stanton D, Wu Y, Weiss RJ, Jaitly N, et al. Tacotron: Towards end-to-end speech synthesis. arXiv [Preprint]. 2017 Mar. arXiv:1703.10135. doi: 10.48550/arXiv.1703.10135
19. Ren Y, Ruan Y, Tan X, Qin T, Zhao S, Zhao Z, et al. Fast-Speech: Fast, robust and controllable text to speech. arXiv [Preprint]. 2019 May. arXiv:1905.09263. doi: 10.48550/arXiv.1905.09263
20. Baas M, Kamper H. StarGAN-ZSVC: Towards zero-shot voice conversion in low-resource contexts. Commun Comput Inf Sci. 2020;1342:69–84. doi: 10.48550/arXiv.2106.00043
21. Yao J, Yang Y, Lei Y, Ning Z, Hu Y, Pan Y, et al. Prompt-VC: Flexible stylistic voice conversion in latent space driven by natural language prompts. arXiv [Preprint]. 2023 Sep. arXiv:2309.09262. doi: 10.48550/arXiv.2309.09262
22. ElevenLabs. Free text to speech online with lifelike AI voices [Internet]. London: ElevenLabs; 2026. Available from: https://elevenlabs.io/text-to-speech
23. Lamothe C, Thoret E, Trapeau R, Giordano B L, Sein J, Takerkart S, et al. Reconstructing voice identity from noninvasive auditory cortex recordings. eLife. 2026 Jan 8:13:RP98047. doi: 10.7554/eLife.98047
24. Casanova E, Davis K, Gölge E, Göknar G, Gulea I, Hart L, et al. XTTS: A massively multilingual zero-shot text-to-speech model. Proc Interspeech 2024. 2024 Sep;4978–4982. doi: 10.48550/arXiv.2406.04904
25. Radford A, Wu J, Child R, Luan D, Amodei D, Sutskever I. Language models are unsupervised multitask learners [Internet]. San Francisco: OpenAI; 2019. Available from: https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf
26. Jaegle A, Gimeno F, Brock A, Zisserman A, Vinyals O, Carreira J. Perceiver: General perception with iterative attention. arXiv [Preprint]. 2021 Mar. arXiv:2103.03206. doi: 10.48550/arXiv.2103.03206
27. Heo HS, Lee B-J, Huh J, Chung JS. Clova baseline system for the VoxCeleb Speaker Recognition Challenge 2020. arXiv [Preprint]. 2020 Sep. arXiv:2009.14153. doi: 10.48550/arXiv.2009.14153
28. Kong J, Kim J, Bae J. HiFi-GAN: Generative adversarial networks for efficient and high fidelity speech synthesis. arXiv [Preprint]. 2020 Oct. arXiv:2010.05646. doi: 10.48550/arXiv.2010.05646
29. Allen DM. Mean square error of prediction as a criterion for selecting variables. Technometrics. 1971;13(3):469–475. doi: 10.2307/1267161
30. Dodson CS, Johnson MK, Schooler JW. The verbal overshadowing effect: Why descriptions impair face recognition. Mem Cognit. 1997 Mar;25:129–139. doi: 10.3758/bf03201107
31. Ardila R, Branson M, Davis K, Henretty M, Kohler M, Meyer J, et al. Common voice: A massively-multilingual speech corpus. Proc 12th Lang Resour Eval Conf. 2020 May;4218–4222. doi: 10.48550/arXiv.1912.06670
32. Kim JW, Salamon J, Li P, Bello JP. CREPE: A convolutional representation for pitch estimation. Proc IEEE Int Conf Acoust Speech Signal Process (ICASS P). 2018 Apr;161–165. doi: 10.48550/arXiv.1802.06182
33. Minibatch AI. Furthest Point Sampling [Internet]. Minibatch AI; 2021. Available from: https://minibatchai.com/2021/08/07/FPS.html
34. Eyben F, Wöllmer M, Schuller B. Opensmile: The munich versatile and fast open-source audio feature extractor. Proc 18th ACM Int Conf Multimed. 2010 Oct;1459–1462. doi: 10.1145/1873951.1874246
35. Facebook. facebook/wav2vec2-large-xlsr-53 [Internet]. Hugging Face; 2021. Available from: https://huggingface.co/facebook/wav2vec2-large-xlsr-53
36. Jadoul Y, de Boer B, Ravignani A. Parselmouth for bioacoustics: Automated acoustic analysis in Python. Bioacoustics. 2024;33(1):1–19. doi: 10.1080/09524622.2023.2259327
37. Derry JR. SciPy. In: Introduction to programming for researchers. Berkeley (CA): Apress; 2026.
38. Agarap AF. Deep learning using Rectified Linear Units (ReLU). arXiv [Preprint]. 2018 Mar. arXiv:1803.08375. doi: 10.48550/arXiv.1803.08375
Informacje: Problems of Forensic Sciences (Z Zagadnień Nauk Sądowych), 2026, 145, s. 45-63
Typ artykułu: Oryginalny artykuł naukowy
Tytuły:
Publikacja: 14.09.2026
Otrzymano: 31.01.2026
Zaakceptowano: 17.06.2026
Status artykułu: Otwarte
Licencja: CC BY-NC-ND 4.0
Udział procentowy autorów:
Korekty artykułu:
-Języki publikacji:
Angielski, PolskiLiczba wyświetleń: 192
Liczba pobrań: 108
