Phencyclidine (PCP), a Schedule II dissociative anesthetic, is no longer prescribed. Recreational use of PCP still exists, often localized to certain metropolitan areas in the United States, including the Baltimore-Washington, D.C. corridor. PCP was detected in 3.07% of the Driving Under the Influence of Drugs (DUID) cases in Virginia and 0.55% of medical examiner cases from 2020 to 2024. For the northern region of Virginia, PCP was detected in 8.08% of the DUID cases and 1.76% of the medical examiner cases from 2020 to 2024. The purpose of this study was to determine the concentration of PCP in a postmortem overdose case where the concentration was reported as present greater than 2000 ng/mL. The subject was a 47-year-old individual who had a history of PCP use. An Agilent Technologies 7890A/5975C Gas Chromatograph-Mass Spectrometer (GC-MS) operated in selected ion-monitoring (SIM) mode was used after PCP was extracted in triplicate using a liquid/liquid alkaline extraction. The mean PCP concentration of the three replicates was approximately 5400 ng/mL. The concentration of PCP in this case is at a much higher concentration than other PCP cases from 2020 to 2024 in Virginia and consistent with literature results for PCP fatalities. After postmortem examination ("foam cone" with no trauma to the body) and review of toxicology results (negative in all analyses except PCP), the cause of death was determined to be "PCP Intoxication." This case provides detailed information on behavior and autopsy findings for a PCP overdose death, where death was due to PCP intoxication as opposed to intoxicated behavior.
Bed-related deaths are rare in forensic practice and are usually associated with falls, restraint-related entrapment, or unsafe infant sleeping environments. Fatal mechanical asphyxia involving lift-up storage beds is an exceptional occurrence, and the few published cases involve adult victims. We report the first pediatric fatality caused by neck entrapment in the closing mechanism of an ottoman-style storage bed. An 8-year-old boy was found prone and unconscious, with his head trapped between the lowered mattress-support frame and the storage compartment below. Facial cyanosis and congestion were observed at the scene. Advanced cardiopulmonary resuscitation restored spontaneous circulation after prolonged cardiac arrest, but the subsequent clinical course was characterized by severe hypoxic-ischemic encephalopathy. Brain death was declared 1 week after the incident. No autopsy was performed. The forensic investigation combined medico-legal assessment, clinical and radiological review, scene examination, and engineering analysis of the bed mechanism. Technical testing excluded a manufacturing defect and demonstrated incorrect assembly of the bed, with accidental interchanging of externally similar components. The available evidence supported the hypothesis of fatal traumatic mechanical asphyxia due to extrinsic cervical compression. Vascular obstruction, particularly impaired jugular venous drainage and reduced cerebral perfusion, was considered the primary mechanism, with possible contributions from carotid compression and reflex neurogenic mechanisms. This case illustrates the importance of integrating circumstantial, clinical, radiological, and technical data. It also demonstrates that ordinary furniture may become lethal when safety-critical components are installed incorrectly, underscoring the need for preventive and safety measures during assembly of lift-up storage beds.
Differential extraction is a core forensic DNA technique separating sperm cells from epithelial cells in sexual assault evidence, facilitating recovery of male DNA profiles from mixed biological samples. Effective physical separation reduces the occurrence of mixed DNA profiles, simplifies interpretation, and conserves analyst time. This process is especially critical for samples with low sperm cell counts, where successful recovery of a probative male profile may be limited. This study evaluated and compared manual and robotic differential extraction methods using 30 mock forensic vaginal swabs with microscopic sperm ratings ranging from 4+ to <1+, representative of sexual assault samples encountered in forensic casework. Swabs were divided into equal portions and extracted in duplicate using both methods by two analysts for reproducibility. Manual and robotic differential extractions were performed using the Qiagen QIAamp® DNA Investigator Kit and standardized "Isolation of Total DNA from Sexual Assault Specimens" protocol for reagent and procedural consistency. Robotic extractions were conducted using the Qiagen QIAcube®. Performance metrics included DNA yield, DNA mixture incidence in epithelial and sperm fractions, and STR profile completeness. Robotic extraction demonstrated procedural consistency and reduced hands-on processing time. However, manual differential extraction produced superior separation performance, with a lower incidence of DNA mixtures in sperm fractions. Samples processed using robotic protocols were significantly more likely to exhibit epithelial carryover into sperm fractions, complicating interpretation and resulting in inconclusive results. In contrast, manual extraction yielded cleaner sperm fractions and improved recovery of single-source male DNA profiles, highlighting the trade-offs between automation and manual separation.
Source camera identification through photo response nonuniformity (PRNU) analysis is a well-established forensic technique, yet the choice of experimental conditions and parameters of the algorithms critically impacts identification reliability. For the first time, this study systematically evaluates 11 denoising filters (linear, nonlinear, and adaptive) for noise pattern extraction in conjunction with six similarity image metrics to establish an optimal framework for PRNU extraction and matching. To measure the PRNU patterns and analyze the identification reliability, a dataset of 4387 light uniform, dark, and natural images of three cameras was created (LDN dataset). Among similarity metrics, normalized standard deviation (NSTD) substantially outperforms Pearson's correlation coefficient (7.6× higher value of the comparison parameter), while peak-to-correlation energy, phase-only correlation, and structural similarity index measure exhibit negligible comparative utility. The Lee filter-originally designed for multiplicative speckle noise reduction-achieves an 8.6× higher comparison parameter over the Wiener filter and 12.1× higher comparison parameter over the median filter. This research establishes a comprehensive experimental framework for increasing the robustness of device-level identification in forensic investigations.
Clear or unpigmented lip gloss, in particular, has become quite popular among people of all genders. Their unpigmented characteristics make them less noticeable at the crime scene; thus, they are less likely to be intentionally removed by the perpetrators. Therefore, in the present study, 20 brands of clear lip gloss were analyzed using attenuated total reflectance-Fourier transform infrared (ATR-FTIR) spectroscopy. Five out of 20 brands exhibited distinct spectral patterns and could be differentiated visually; however, the remaining 16 brands were grouped into five categories. Trend analysis was performed using principal component analysis (PCA), and samples within each group were found to be either clustered or located close to each other on the PCA plot. Further, for the discrimination of lip gloss samples, the support vector machine (SVM) method was employed. 97.5% of training as well as cross-validation accuracy was achieved. To ensure the reliability of the model, external validation and a blind test were also performed, which yielded 90% and 100% prediction accuracy, respectively. Moreover, substrate study on various porous and non-porous substrates in the time interval of 1 day, 7 days, and 15 days was also conducted. Non-porous substrates (ceramic, steel, plastic) did not affect the classification of samples and allowed for accurate sample classification (100%) for up to 15 days. However, samples on porous substrates were poorly classified. Therefore, ATR-FTIR spectroscopy and chemometric classification can potentially be used in real-world cases where questioned samples could be linked to their source and offer investigative leads.
With the nascent reliance on unmanned aerial systems as a means of surveillance, security, and monitoring, it is essential to ensure that video data legitimacy is assured. Object-based Copy-Move Forgery (CMF), in which an object is either copied or moved in the frame or across a frame to alter the visual description, is one of the major threats to video integrity. Such manipulations are a problem for traditional detection methods because object motion, occlusion, and irregular background are complex. This research presents an interpretable and robust deep learning architecture to identify such forgeries, that is, a combination of transfer learning with a new Multi-Aspect Fossa Graph Transformer enhanced with Shuffle Attention (MAFGTN-SA). To achieve this, a better generative adversarial network, which is built upon GoogLeNet, is used to extract deep spatial-semantic features using normalized grayscale video frames. These characteristics are then represented by MAFGTN-SA to represent complex object relationships between the frames and spatial inconsistencies to accurately classify genuine and tampered frames. In order to increase interpretability, the model incorporates Local Interpretable Model-Agnostic Explanations (LIME) for saliency-based visualization of tampered regions. Experimental evaluations on three benchmark datasets, SULFA, GRIP, VTD, CG-1050 v2.0, and COVERAGE, validate the effectiveness of the proposed method, attaining an accuracy of 98.73%, 96.78%, 98.45%, 97.92%, and 97.36%, respectively, and with high F1-scores of 98.73%, 98.5%, 97.98%, 97.41%, and 96.93%. The results highlight the framework's superior detection capability, reliability, and explainability across diverse video forgery scenarios.
Sinus of Valsalva aneurysm (SVA) is a rare cardiac anomaly, with autopsy-confirmed fatal cases remaining uncommon. This report presents a rare case of sudden death in a 13-year-old male adolescent caused by rupture of a noncoronary SVA into the right atrium, resulting in acute circulatory dysfunction, with vomiting as the only initial symptom. The cause of death was confirmed through comprehensive forensic autopsy, histological examination, and toxicological analysis. Additionally, a review of relevant autopsy case reports published in the past two decades on PubMed and Web of Science identified 13 cases of SVA-related mortality. We summarized the anatomical origin of aneurysms, gender distribution, clinical manifestations, rupture-associated fatal mechanisms, and pathological features observed in these cases. The findings indicated that in SVA-related fatal cases, the most common rupture sites were the right coronary sinus, left coronary sinus, and noncoronary sinus. The rupture typically occurred into the atrium or pericardium. Chest pain and vomiting were frequently reported as prodromal symptoms. Imaging techniques such as echocardiography and computed tomography were useful in aiding diagnosis. Cardiac pathology in these cases was often accompanied by additional abnormalities, including cardiomegaly, atherosclerosis, inflammation, or congenital defects. This report, in conjunction with the literature review, provides important reference material for forensic autopsy.
Skeletal age estimation is crucial to medicolegal work, aiding subadult identification, and living age assessments in legal contexts. This study aims to document the ontogeny of knee skeletal indicators on post-mortem computed tomography (PMCT), and quantify interpopulation variation between modern American and Australian children. PMCT data from 402 Australian subadults from Victoria, Australia (228 males, 174 females), and 445 North American subadults from New Mexico (233 males, 202 females) were retrospectively accessed. Thin-slice data were visualized as multiplanar reconstructions, and an ordinal scoring system applied to 11 indicators on the femur, tibia, and fibula. High repeatability was exhibited by excellent intra-rater (ICC = 0.99) and good inter-rater (ICC = 0.89) agreement. Transition analysis produced maximum likelihood maturation estimates, with age parameters derived from a multivariable Bayesian model using 68-95% credible intervals. Active knee fusion occurred in Australian females aged 10-16 and males 12-18, with complete fusion of all sites no earlier than 14 and 16 years. North American females showed slightly later fusion (11-15 years), consistent with males (11-16 years). Sexual dimorphism was consistent in both subpopulations, and interpopulation differences were especially prominent in later adolescence to early adulthood for the epiphyses for males, but were not identified for most morphological indicators. Recalibrated, normative PMCT knee development standards are presented for Australian and North American subadults, identifying four indicators especially informative for age estimation across developmental profiles derived from a Bayesian cluster model. Our findings support use of population-specific standards, particularly in later fusion stages after 15-16 years, to improve accuracy in forensic and clinical applications.
Three US government agencies, the Drug Enforcement Administration (DEA), Customs and Border Protection (CBP), and Food and Drug Administration (FDA), established a collaborative program to leverage each agency's expertise and capabilities to characterize illicit fentanyl-containing tablets and powders collected at US ports of entry (POEs) for attribution and sourcing purposes. The Intelligence National Threat Response-El Paso Illicit Drug (INTREPID) program employed a proof-of-concept pilot study that combined physical and chemical analytical techniques with machine learning to establish comprehensive sample profiles. Samples were analyzed using complementary techniques including color imaging, three-dimensional surface analysis, and multi-technique chemical profiling (Raman, FT-IR, GC-MS, LC-MS, ICP-MS, 1H NMR). Machine learning models were employed solely as screening tools to prioritize samples for detailed human analysis; all conclusions were based on direct analytical comparison, not on machine learning results. Physical analyses plausibly linked fentanyl tablet seizures from two different ports of entry collected 8 weeks apart based on consistent color characteristics and three-dimensional debossing patterns. Chemical analyses plausibly linked fentanyl powder seizures from the same port of entry collected 3 h apart with identical qualitative profiles and minimal quantitative variation (16.7% concentration error). Although alternative explanations for the similarities observed between these sample pairs cannot be excluded, these findings demonstrate the feasibility of using combined physical and chemical data within a forensic intelligence framework to generate actionable leads for law enforcement investigations of illicit fentanyl trafficking networks.
The availability of highly sensitive multiplex kits has resulted in the more common observation of low-level and more complex (multiple contributors) DNA profiles within forensic casework. With the emergence of probabilistic genotyping software these complex DNA mixture profiles can now be interpreted. This has led to a shift in questioning within courtroom settings, where the number of contributors (NoC) assigned by the forensic DNA analyst is now often scrutinized. This study aims to assess the likelihood ratio (LR) behavior for the under- and over-assignment of a range of complex mixtures, including six-person mixtures, using the probabilistic genotyping software STRmix™. The mixtures were analyzed using FaSTR™ DNA and a NoC was "blindly" assigned by an analyst (i.e., without prior knowledge of the experimental design). LRs obtained for true contributors and non-contributors, under different NoC assignments, were compared and showed that overestimating NoC may lead to weak adventitious matches, while underestimating can lead to the false exclusions of true contributors. Little effect was observed on the LR assigned to the major contributors to these complex mixtures. Moreover, as the profile complexity increased (higher order mixtures), the impact on the LR for true contributors, when underestimating NoC by one, diminished, with only 35 of the 72 comparisons (~49%) resulting in a smaller LR for the under-assignment. It is paramount for DNA analysts to understand the LR behavior especially when questioned on the witness stand regarding their NoC assignment, as there is no ability to reassess or reinterpret the evidence in real-time.
Decoded pulse-code modulation (PCM) stream hashing is used for forensic content verification, yet its reliability depends on decoder output stability. This study examines whether the iTunes Seamless Playback (iTunSMPB) metadata field affects decoded PCM alignment and stream hash reproducibility in Advanced Audio Coding-Low Complexity (AAC-LC) recordings generated by iOS devices. A controlled dataset of 280 recordings was collected from seven iPhone models across multiple iOS versions. Eleven AAC-LC file pairs containing clean and tagged versions were analyzed to assess tag insertion conditions, decoded PCM alignment, signal integrity, system database behavior, and the effectiveness of metadata-only versus container-rewrite remediation workflows. The presence of the iTunSMPB field corresponded to reproducible reductions in leading zero-amplitude sample padding and associated SHA-256 decoded PCM stream hash mismatches in 8 of 11 AAC-LC pairs. Waveform, spectral, and quantitative integrity measures (LTASS, PCC, MQD) remained unchanged. Apple Lossless Audio Codec (ALAC) recordings were unaffected regardless of tag presence. Metadata-only tag removal restored stream hash alignment for all affected files without altering container structure, whereas FFmpeg-based container rewriting restored 9 of 11 files and introduced structural changes in all cases. These findings demonstrate that decoded PCM stream hash mismatches can occur without audio editing and should not be interpreted as evidence of tampering without considering metadata state, container branding, and decoder behavior. Future work should expand testing to non-iOS AAC encoders, alternative decoders, and additional container variants, and develop decoder-independent PCM hashing methods robust to metadata-associated alignment effects.
Most homicides are firearm-perpetrated, but in sexual homicide, offenders are more likely to utilize personal weapons (e.g., hands and feet), contact weapons (e.g., blunt object), or edged weapons (e.g., knife) to comport with their fantasy life, motivational states, and psychopathology. Despite these conceptual designations, epidemiological data suggest roughly 17% to 31% of sexual homicide offenders kill their victim with a firearm, yet we know surprisingly little about these offenders' criminal careers, psychopathology, and offense conduct. Drawing on a near-census of condemned capital murderers from a Western US state, we found just 8% are firearm sexual homicide offenders. Relative to sexual homicide offenders who did not use a firearm and other capital murderers, firearm sexual homicide offenders have more murder convictions, total arrests, and total convictions, are more likely Black, are more likely to rob and execute their victims, and are marginally more sexually sadistic. Offenders who kidnap their victims are 6.5 times more likely to be firearm sexual homicide offenders. Our findings add empirical granularity about the criminal history, psychological profile, and offense conduct of those who shoot their victims during sexual homicide.
Forensic anthropology has garnered greater attention in the academic and medicolegal communities with more practitioners embedded in Medical Examiner's and Coroner's offices. Practitioners have varied roles and responsibilities, leaving few opportunities to compare the nature, breadth, and outcomes of their casework. Sharing data provides the foundations upon which resource distribution, planning, and needs assessment can be based, providing essential information for training the next generation of professionals. This study reports on casework conducted by three forensic anthropology practitioners across two Midwest Medical Examiner's jurisdictions. Between 2013 and 2022, forensic anthropologists working as consultants at the Cook County Medical Examiner's Office (CCMEO) completed 140 cases, with the St. Louis Medical Examiner's Office (SLMEO) having an embedded anthropologist completing 288 cases. The greatest number of cases were recorded in 2022 for both agencies and entered the CCMEO in June and July, and the SLMEO between October and December. Recovery of decomposed and fully skeletonized decedents from outdoor surface contexts (CCMEO) and indoor scenes (SLMEO) was most common. Biological profile (CCMEO) and radiographic identification (SLMEO) were the most requested types of evaluation. Differences exist in timing, analysis, and report submission. Variation stems from contractual differences between offices, environmental contexts, and the type of consult requested. Greater discussion and comparison of the range of services and expertise that forensic anthropologists provide broadens the understanding of the field, informs our stakeholders, allows for the formulation of best practices, and serves as data upon which decisions regarding protocol, funding, resources, and need can be based.
Many readily available online videos and instructional manuals provide information on how to manufacture aluminum (Al) powder, a component in pyrotechnics and improvised explosive devices (IEDs) using common household materials. Given the ease in acquiring and producing Al powders, it is common to observe Al powder in IEDs. Thus, there is a need to characterize Al powders, in a forensic context, to determine whether there are signatures that could be used for sample differentiation and placing potential constraints on the method of Al powder production. In this study, samples of Al powder that were derived from industrial products (including end-use commercial products) and those made using amateur methods were analyzed. The powders were analyzed using a combination of morphology (viewed with light and scanning electron microscopy (SEM)) and elemental profiling (using energy dispersive X-ray spectroscopy (EDS)). The results demonstrate that Al powder manufactured by industrial processes can be readily differentiated from amateur methods. The gross microscopic morphology of the particles provided the basis for the recognition of several different particle types/classes. These types provide insight into their method of production. Of particular note is the ability of morphological characteristics to differentiate between flake-type powders, which are made by both industrial and amateur methods. This differentiation is primarily due to significant differences in the particle size distribution and the particle edge textures.
Smartphones provide crucial but duplicate information in forensic investigations. The major deduplication tools designed for the cloud ignore constraints on phone storage, power, data fragmentation, and dynamic handling, leading to processing delays and limiting their efficient use. It also affects evidence integrity and increases the potential risk of data leakage. This paper proposes a data deduplication technique for resource-constrained Android devices with limited computing power and memory. The use of the Blake2b-Hash Algorithm in the proposed deduplication algorithm reduces overhead and enables dynamic memory allocation. Multiple strategies were proposed and tested using a wide range of Smartphones. The proposed algorithm is validated based on file analysis time, device storage and memory utilization, and storage overhead. The paper also assesses the hash function's cryptographic robustness. The experimental analysis demonstrates that the proposed method achieved a 45.90% processing time to identify 29.52% duplicate files. A 48.55% reduction in storage space was achieved, and battery consumption was reduced to 81.15% compared to MD5 and SHA224. It achieved up to 52.71% RAM optimization, along with only 0.0033% storage overhead to maintain the metadata of deleted files for forensic purposes. This proves the effectiveness and reliability of the proposed solutions in facilitating digital forensic analysis.
Photo paper is frequently encountered in forensic investigations; however, latent fingermark development remains challenging because its surface characteristics vary according to the manufacturing process and coating structure. This study investigated suitable methods for developing latent fingermarks on commercially available photo papers. Seventeen commercially available photo papers representing two manufacturing types, resin-coated papers (RCPs) and cast-coated papers (CCPs), were examined. Latent fingermarks were developed using 1,2-indanedione/zinc chloride (IND/Zn) under different heat development conditions, either alone or in combination with cyanoacrylate (CA) fuming. The results showed clear differences between RCPs and CCPs. For RCPs, the CA-masking method (CA fuming on the nonprinting surface while masking the printing surface, followed by IND/Zn treatment on the printing surface) was the most effective approach, whereas for CCPs, IND/Zn treatment alone was optimal for both surfaces, with heating at 110°C for 60 s recommended for both paper types. In contrast, IND/Zn alone was effective for both surfaces of CCPs. The proposed methods were also effective for naturally aged fingermarks. These findings provide practical guidance for selecting fingermark development procedures according to the manufacturing type and surface characteristics of photo paper.
A substantial proportion of forensic evidence consists of skin or contact traces, commonly referred to as touch DNA. The collection of such biological material remains challenging due to its invisibility and low abundance. Further, depending on substrate type and surface properties, different recovery methods are employed, with adhesive tapes and stamps commonly applied to textile evidence. Although tape lifting is routine in many jurisdictions, limited data exist on the relative efficiency of adhesive materials, the optimal number of lifts, and their comparative performance in touch DNA recovery from textiles. To address the gap, this study evaluated four commercially available adhesive tapes/stamps for touch DNA collection from fabrics. A high DNA-shedding volunteer deposited touch DNA on cotton, denim, and, unexamined fleece in replicates (n = 360) using DNA Tape, Scene Safe Fast™ Minitapes, Force-Forex Adhesive Stamp, and DNA Stub Recording Stamp, as well as standard and increased lysis buffer volumes. No statistically significant differences were observed in DNA yield or STR profile quality among tapes, though substrate type notably influenced DNA recovery, and handling differences among products were noticed. Touch DNA deposition on denim showed high intraindividual variability. Increased lysis buffer volume did not alter extraction efficiency. Considering both performance and usability, the DNA Tape remained the preferred option in this DNA laboratory, while the DNA Stub Recording Stamp represented a practical alternative for on-site police collection.
In the medicolegal context, GSR analysis plays a key role in clarifying the dynamics of firearm-related deaths, including lesion identification and firing distance determination. However, postmortem alterations of the body, such as thermal exposure, may influence the analysis of lesions and complicate interpretation. This study aimed to evaluate the persistence and modifications of GSR under different firing distances and thermal treatments using Scanning Electron Microscopy equipped with Energy-Dispersive X-ray Spectroscopy (SEM-EDS) and Inductively Coupled Plasma Optical Emission Spectroscopy (ICP-OES). Ballistic tests were conducted on calcium-magnesium silicate substrates using a self-reloading 9 × 21 mm pistol at five barrel-to-target distances (5-100 cm). To simulate fire exposure, a subset of the targets was heat-treated in a muffle furnace for 30 min at 250, 750, and 1000°C. SEM-EDS and ICP-OES analyses were then performed, focusing on morphology and chemical composition. GSR remained detectable under all experimental conditions. Both firing distance and thermal exposure affected gunshot residues, leading to changes in morphology, particle size, and elemental composition. An antagonistic interaction between these variables on GSR composition was identified, indicating that, depending on the temperature exposure, metal concentrations may vary with firing distance. Among the metallic elements, lead showed the strongest response, as explained by its low melting point. Notably, at 1000°C, no particles fulfilled the ASTM E1588-25 criterion for characteristic GSR (Pb, Ba, and Sb), rendering the trimetallic population undetectable. These results highlight changes in GSR characteristics compared to standard criteria, suggesting the need to adapt GSR protocols in cases of burned remains.
Bee sting deaths represent an uncommon but important forensic problem because death may follow either anaphylaxis after only a few stings or toxic envenomation after massive attacks, while autopsy findings may be subtle and nonspecific. We present three fatal medicolegal autopsy cases from Lucknow, India involving rapid collapse after honey bee attacks, with sting burdens ranging from limited exposure to mass envenomation. Case 1 showed multiple stings over exposed upper body parts with facial edema and generalized congestion but no significant upper airway or pulmonary edema. Case 2 showed multiple posterior body stings, tracheobronchial froth, and markedly heavy edematous lungs. Case 3 showed an estimated 200 stings concentrated over the face and upper body, severe epiglottic and laryngeal edema with luminal narrowing, tracheobronchial froth, and marked pulmonary edema. Together, these cases demonstrate the morphological spectrum of fatal bee sting reactions and underline the importance of integrating scene findings, witness history, sting distribution, and focused airway examination when ancillary tests are unavailable.
With the rapid advancement of generative artificial intelligence, deepfake speech has emerged as a significant threat to digital audio authenticity, posing challenges for forensic analysis and legal applications. In this study, we propose EMBNet, a task-oriented deepfake speech detection framework that integrates efficient channel attention (ECA) with a multi-scale bottleneck to enhance the representation of subtle acoustic anomalies. The ECA module adaptively emphasizes critical feature channels, while the multi-scale bottleneck captures local and hierarchical spoofing traces across multiple time-frequency resolutions. The proposed framework effectively balances fine-grained local detail modeling with global hierarchical representation, improving the detection of weakly manifested spoofing patterns. Extensive experiments on the ASVspoof 2019 Logical Access dataset demonstrate that EMBNet significantly outperforms existing baseline and state-of-the-art methods, achieving an EER of 2.67%, an AUC of 97.32%, and an F1-score of 97.28%. Ablation studies further confirm the complementary contributions of the ECA and multi-scale modules to overall performance. The proposed approach demonstrates promising performance for forensic audio analysis and may support forensic audio authenticity assessment by improving the discrimination between genuine and manipulated speech.
