Paper deep dive

Teffic-Audio: Tell Fact from Fiction

Wan Lin, Li Wang, Jindong Wang, Kunyu Feng, Zhizheng Wu

Year: 2026Venue: arXiv preprintArea: cs.SDType: PreprintEmbeddings: 73

Intelligence

Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 93%

Last extracted: 8/3/2026, 10:06:26 AM

Summary

The paper introduces Teffic-Audio, a general speech deepfake detection system that achieves state-of-the-art performance on the Speech-DF-Arena benchmark with a pooled EER of 1.454%. The system utilizes a Conformer-based encoder, multi-head attentive statistics pooling, and a binary classifier. Its success is attributed to a training recipe involving multi-source data integration, attack- and source-balanced sampling, and diverse audio augmentation, rather than architectural complexity.

Entities (10)

Teffic-Audio · model · 100%EER · metric · 99%Speech DF Arena · benchmark · 98%Speech-DF-Arena · dataset · 98%Conformer · modelcomponent · 95%Multi-head Attentive Statistics Pooling · modelcomponent · 95%ADD · dataset · 90%ASVspoof · dataset · 90%RawBoost · technique · 90%W2V-BERT 2.0 · model · 90%

Relation Signals (9)

Teffic-Audio → achievesmetric → EER

confidence 98% · achieves a pooled EER of 1.454%

Teffic-Audio → evaluatedon → Speech DF Arena

confidence 98% · Teffic-Audio achieves a pooled EER of 1.454% on the 14 test sets of Speech-DF-Arena

Teffic-Audio → outperforms → all currently public systems

confidence 95% · outperforming all currently public systems on the leaderboard.

Teffic-Audio → usescomponent → Conformer

confidence 95% · Teffic-Audio adopts a straightforward detector architecture consisting of a Conformer-based speech encoder

Teffic-Audio → usescomponent → Multi-head Attentive Statistics Pooling

confidence 95% · multi-head attentive statistics pooling (MHASP) to obtain a fixed-dimensional utterance-level representation.

Speech DF Arena → includes → ASVspoof

confidence 90% · The evaluation sets cover the ASVspoof series

Speech DF Arena → includes → ADD

confidence 90% · the ADD challenge series

Teffic-Audio → trainedwith → multi-source data

confidence 90% · integrates multi-source data, attack- and source-balanced sampling, and diverse audio augmentation.

→ →

Cypher Suggestions (0)

No Cypher suggestions yet.

Abstract

Abstract:Speech deepfake detection has expanded in scope with increasingly heterogeneous spoofing mechanisms, including speech synthesis, voice conversion, vocoder reconstruction, and neural-codec resynthesis. The resulting spoofing artifacts can be further shaped by variability in source speech, recording environments, and transmission channels. This variability makes robust generalization across heterogeneous conditions a central requirement for practical detection systems. This report presents Teffic-Audio, a general speech deepfake detection system designed for comprehensive evaluation environment. Teffic-Audio adopts a straightforward detector architecture consisting of a Conformer-based speech encoder, multi-head attentive statistics pooling, and a binary classifier. Rather than relying on additional architectural complexity, the system improves generalization through its training recipe, which integrates multi-source data, attack- and source-balanced sampling, and diverse audio augmentation. Trained only with open-source data, Teffic-Audio achieves a pooled EER of 1.454% on the 14 test sets of Speech-DF-Arena, outperforming all currently public systems on the leaderboard. It also obtains the lowest EER on five individual test sets and shows a favorable performance-complexity trade-off compared with larger leading systems. Overall, Teffic-Audio provides a strong and practical reference system for general speech deepfake detection.

PDF

Open source PDF →Open local PDF →

Trouble viewing inline? Open PDF directly →

Full Text

71,465 characters extracted from source content.

Expand or collapse full text

Teffic-Audio: Tell Fact from Fiction Amphion Team Abstract Speech deepfake detection has expanded in scope with increasingly heterogeneous spoofing mechanisms, including speech synthesis, voice conversion, vocoder reconstruction, and neural-codec resynthesis. The resulting spoofing artifacts can be further shaped by variability in source speech, recording environments, and transmission channels. This variability makes robust generalization across heterogeneous conditions a central requirement for practical detection systems. This report presents Teffic-Audio, a general speech deepfake detection system designed for comprehensive evaluation environment. Teffic-Audio adopts a straightforward detector architecture consisting of a Conformer-based speech encoder, multi-head attentive statistics pooling, and a binary classifier. Rather than relying on additional architectural complexity, the system improves generalization through its training recipe, which integrates multi-source data, attack- and source-balanced sampling, and diverse audio augmentation. Trained only with open-source data, Teffic-Audio achieves a pooled EER of 1.454% on the 14 test sets of Speech-DF-Arena, outperforming all currently public systems on the leaderboard. It also obtains the lowest EER on five individual test sets and shows a favorable performance–complexity trade-off compared with larger leading systems. Overall, Teffic-Audio provides a strong and practical reference system for general speech deepfake detection. Keywords Speech Deepfake Detection ⋅· Speech-DF-Arena ⋅· Generalization 1 Introduction Recent advances in speech generation Ju et al. (2024); Du et al. (2024b); Chen et al. (2024, 2025) continue to broaden the scope of speech deepfake detection. The sources of spoofed speech are no longer confined to a small set of speech synthesis or voice conversion methods Wu et al. (2015); Wang et al. (2020). They now cover generation systems that differ in development stage, technical family, and processing pipeline Müller et al. (2024b); Li et al. (2025a). Recent studies also treat vocoder-reconstructed audio Sun et al. (2023); Frank and Schönherr (2021) and audio resynthesized after neural codec compression and decoding Xie et al. (2025) as targets in spoofing detection or spoofing-related evaluation, which further expands the boundary of the detection problem. The artifacts produced by different generation mechanisms are often not consistent Müller et al. (2024a). Even when spoofing traces appear similar, their observable forms can vary with the source of bonafide speech, recording conditions, channel transmission, and compression codecs Chandra et al. (2025); Shi et al. (2025). The central goal of speech deepfake detection therefore shifts from identifying specific spoofing methods to maintaining stable and general discriminative ability across diverse bonafide speech sources, spoofing mechanisms, and propagation conditions Müller et al. (2024a); Dowerah et al. (2026). This change in task scope is also reflected in the evolution of evaluation benchmarks, which move from specific anti-spoofing scenarios toward comprehensive generalization assessment. The early ASVspoof series Wu et al. (2015); Wang et al. (2020); Yamagishi et al. (2021) mainly formulates evaluation tasks around typical attack scenarios such as logical access and physical access. As speech spoofing techniques and application environments continue to change, the ADD challenges Yi et al. (2022, 2023) and independent evaluation sets such as In-the-Wild Müller et al. (2022) extend the evaluation coverage to more complex audio conditions, including low-quality speech and real-world noise. ASVspoof 5 Wang et al. (2024) further organizes evaluation over larger-scale speech sources, attack types, and adversarial conditions, imposing stronger requirements on the generalization and robustness of detectors. Nevertheless, these evaluations remain largely tied to their own data construction strategies and protocol designs. In contrast to benchmarks centered on a single evaluation source, Speech-DF-Arena Dowerah et al. (2026)111https://huggingface.co/spaces/Speech-Arena-2025/Speech-DF-Arena integrates multiple representative test sets under a unified protocol and metric system. This design makes cross-dataset stability directly comparable and provides a more suitable comprehensive environment for evaluating general speech deepfake detectors. Existing studies improve speech deepfake detection systems along several axes to enhance robustness and generalization. Earlier methods mainly strengthen the ability of detectors to capture spoofing cues through spectral feature modeling Wu et al. (2020); Li et al. (2021), end-to-end waveform modeling Tak et al. (2021b), or spectro-temporal relation modeling Jung et al. (2022). With the development of self-supervised speech models, pretrained encoders such as wav2vec 2.0 Baevski et al. (2020), WavLM Chen et al. (2022), and XLS-R Babu et al. (2021) have become important bases for strong detectors because they provide richer acoustic and speech representations for downstream detection. Building on these encoders, recent methods further exploit discriminative information across different representation levels, temporal scales, and acoustic conditions through multi-layer feature fusion Zhang et al. (2024); Wang et al. (2025), attentive aggregation Truong et al. (2024); Li et al. (2025b), sequence modeling Tran et al. (2025), or expert routing Negroni et al. (2025); Pan et al. (2025). Beyond model and representation design, data factors also receive increasing attention. Augmentation methods Cohen et al. (2022) such as RawBoost Tak et al. (2022a) aim to improve robustness to changes in propagation conditions by simulating channel, codec, and nonlinear distortions. Ge et al. Ge et al. (2025) emphasize the importance of large-scale datasets for adapting pretrained SSL representations. Data-centric methods such as DOSS Huang et al. (2025c) further discuss the influence of generator diversity and heterogeneous data mixing on cross-dataset generalization. These studies demonstrate that detection performance is shaped by multiple factors. Nevertheless, existing methods are either mainly validated on specific benchmarks or limited cross-domain settings Yang et al. (2025); Huang et al. (2025a, 2026), or have not yet shown sufficiently strong overall performance in a comprehensive multi-dataset evaluation environment Kulkarni et al. (2026); Laakkonen et al. (2026). As a result, there is still limited evidence on how to build a general speech deepfake detector that can adapt to heterogeneous evaluation scenarios. Under this background, this report introduces Teffic-Audio222Demo page: https://tefficlabs.com/teffic-audio, a general speech deepfake detection system designed for the comprehensive evaluation environment. The system adopts a standard detector architecture, consisting of a Conformer-based Gulati et al. (2020) speech encoder, multi-head attentive statistics pooling Okabe et al. (2018), and a binary classifier. On this basis, the system focuses on the design of the training procedure, including multi-source data construction, attack- and source-balanced sampling, and diverse audio augmentation. These designs jointly contribute to constructing a more effective training distribution. Our system is trained only with open-source data and achieves a pooled EER of 1.454% on the 14 test sets of Speech-DF-Arena; when compared with all currently public systems, this result ranks first. Meanwhile, Teffic-Audio achieves the current lowest EER on 5 individual test sets and shows a favorable performance–complexity trade-off when compared with leading systems on the current public leaderboard. These results provide direct evidence for building a general speech deepfake detector: a simple detector architecture can achieve strong generalization when supported by a well-designed training distribution, without necessarily relying on increasingly complex detection architectures. Further ablation studies show that diverse audio augmentation is a key factor in improving the system’s performance in complex and difficult scenarios. In addition, the choices of encoder backbone and pooling layer both have a significant impact on the final system performance. Notably, we also find that a system variant with only 4 Conformer blocks can still achieve a pooled EER of 3.346%, indicating that the system can maintain strong cross-dataset generalization with substantially reduced encoder depth. Figure 1: Overview of the Teffic-Audio system architecture and the waveform-to-score training pipeline. 2 System Overview The system targets utterance-level speech deepfake detection. Given an input speech utterance, the model outputs a detection score that represents the likelihood of the utterance being spoofed speech. The overall architecture consists of a SSL-initialized speech encoder, a pooling layer, and a binary classifier, forming an end-to-end discriminative model. The system architecture is illustrated on the right side of Figure 1. The speech encoder is initialized from w2v-BERT 2.0 Barrault et al. (2023). Its backbone comprises a CNN-based feature extractor followed by 24 Conformer blocks Gulati et al. (2020), each with a model dimension of 1024. The encoder converts the input waveform into contextualized frame-level speech representations. On top of the frame-level representations, the system adopts multi-head attentive statistics pooling (MHASP) Okabe et al. (2018) to obtain a fixed-dimensional utterance-level representation. MHASP uses 4 attention heads to compute attention weights along the temporal dimension and estimates the weighted mean and weighted standard deviation accordingly. The statistics from all heads are concatenated to form a 2048-dimensional pooling representation, which is then projected by an MLP with hidden dimensions of 2048→1536→10242048→ 1536→ 1024. This design aggregates discriminative information along the temporal dimension from multiple attention views. Finally, the system uses an MLP classifier with a hidden dimension of 512 to map the 1024-dimensional global representation to binary classification logits. After sigmoid activation, the posterior probability of spoofed speech is used as the detection score. During training, the model is optimized with the binary cross-entropy (BCE) loss: ℒBCE=−1N∑i=1N[yilog⁡pi+(1−yi)log⁡(1−pi)],L_BCE=- 1N _i=1^N [y_i p_i+(1-y_i) (1-p_i) ], (1) where yi∈0,1y_i∈0,1 denotes the binary label of the i-th sample, and pip_i denotes the posterior probability that the sample is predicted as spoofed speech. This objective is applied directly to the complete detection path, enabling the speech encoder, pooling layer, and classifier to be jointly optimized in an end-to-end manner. During inference, this probability is directly used as the final score without additional post-processing. 3 Training Recipe The training recipe is a central component of the system design. In a comprehensive evaluation environment, the main challenge arises from distributional heterogeneity caused jointly by different spoofing mechanisms, bonafide speech conditions, and propagation chains. This section describes how the training procedure is designed to obtain a more comprehensive supervision distribution. As illustrated on the left side of Figure 1, the system is developed from three aspects: Multi-source Training Corpus extends the source coverage of bonafide and spoofed speech, Attack- and Source-Balanced Data Sampling adjusts the gradient contributions of different spoofing mechanisms and bonafide speech sources, and Diverse Audio Augmentation increases the observable variation of training samples under different acoustic environments and transmission conditions. 3.1 Multi-source Training Corpus A single dataset usually reflects only limited speech sources, collection conditions, and spoofing methods, making it insufficient to support cross-domain generalization in comprehensive evaluation. To reduce source bias introduced by the training corpus itself, the system integrates multi-source open-source speech data. Overall, the training corpus consists of two parts. The first part is the speech deepfake corpus, which contains public speech deepfake detection datasets and covers major spoofing mechanisms, including text-to-speech (TTS), voice conversion (VC), neural vocoder reconstruction (NV), and neural codec resynthesis (NC). The second part is the supplementary bonafide corpus, which consists of additional real speech datasets and is used to expand the coverage of bonafide speech in terms of source, speaker, language, and recording condition. Table 1 summarizes the main sources, languages, sample sizes, and generator metadata of the training corpus. Table 1: Multi-source training corpus used in the proposed system. # Bonafide and # Spoofed denote the numbers of real and spoofed training utterances. Generator Types denotes coarse spoofing mechanisms, text-to-speech (TTS), voice conversion (VC), neural vocoder reconstruction (NV), and neural codec resynthesis (NC). # Generator is the number of distinct spoof generators in metadata when annotated. Dataset Year Language # Bonafide # Spoofed Generator Types # Generator Speech Deepfake Corpus ASVspoof2015 (Wu et al., 2015) 2015 EN 9,404 184,000 TTS, VC 10 ASVspoof2019LA (Wang et al., 2020) 2019 EN 2,580 22,800 TTS, VC 6 ASVspoof5 (Wang et al., 2024) 2024 EN 18,797 163,560 TTS, VC 8 ADD2022 (Yi et al., 2022) 2022 ZH 3,312 24,772 TTS, VC – ADD2023 Track1 (Yi et al., 2023) 2023 ZH – 24,072 TTS, VC – FakeOrReal (Reimao and Tzerpos, 2019) 2019 EN 26,900 26,900 TTS – SpoofCeleb (Jung et al., 2025) 2024 EN 230,948 2,309,473 TTS 10 ReplayDF (Müller et al., 2025) 2025 MULTI 26,160 26,160 TTS 4 DFADD (Du et al., 2024a) 2024 EN 44,455 163,500 TTS 5 MLAAD (Müller et al., 2024b) 2024 MULTI – 172,393 TTS 58 LibriSeVoc (Sun et al., 2023) 2023 EN 13,201 79,206 NV 6 SpeechFake (Huang et al., 2025b) 2025 EN, ZH 75,708 629,154 TTS, VC, NV 30 Wavefake (Frank and Schönherr, 2021) 2021 EN, JA 9,170 117,985 NV 6 CodecFake (Xie et al., 2025) 2024 EN, ZH 105,821 634,926 NC 7 Supplementary Bonafide Corpus LibriSpeech (Panayotov et al., 2015) 2015 EN 281,241 – – – AISHELL3 (Shi et al., 2020) 2020 ZH 88,035 – – – GigaSpeech (Chen et al., 2021) 2021 EN 885,397 – – – CNCeleb (Fan et al., 2020) 2019 ZH 524,787 – – – CommonVoice (Ardila et al., 2020) 2020 MULTI 167,571 – – – • Speech Deepfake Corpus. The speech deepfake corpus brings together speech spoofing data from different development stages. Early data mainly come from ASVspoof2015 and ASVspoof2019LA, which are used to cover typical TTS and VC attacks in the logical access scenario. The training corpus then incorporates datasets such as ADD2022, ADD2023 Track1, FakeOrReal, SpoofCeleb, ReplayDF, and ASVspoof5 to supplement samples from more complex challenge protocols and noisy conditions. Furthermore, the system introduces DFADD, MLAAD , LibriSeVoc, SpeechFake, Wavefake, and CodecFake to provide spoofing-related samples from more recent or more specific generation pipelines. These data mainly involve diffusion / flow-matching TTS, multilingual TTS, neural vocoder generation or reconstruction, and neural codec reconstruction, thereby extending the coverage of the training corpus to new generation and reconstruction pipelines. • Supplementary Bonafide Corpus. In addition to the bonafide speech included in the speech deepfake corpus, this system further introduces a supplementary bonafide corpus to enhance the distributional coverage of real speech. Recent studies Kwok et al. (2025) show that models trained on existing deepfake detection data can suffer performance degradation under complex real-speech conditions, and some bonafide speech may be misclassified as spoofed speech due to differences in noise, recording environment, or speaking style. Based on this observation, we explicitly increase the source diversity of bonafide speech during training to reduce the dependence of the model on limited real-speech conditions. Specifically, the supplementary corpus is divided into two categories. The first category contains relatively standard read-speech recordings, including LibriSpeech and AISHELL3, which are used to supplement clean and large-scale bonafide speech. The second category contains real speech data from more open sources and more complex scenarios, including GigaSpeech, CommonVoice, and CNCeleb, which are used to introduce stronger variations in speaker, language, content, and recording condition. 3.2 Attack- and Source-Balanced Data Sampling As shown in Table 1, different training data sources contribute substantially different numbers of utterances, and the number of attack generators covered on the spoofed speech side also varies. If all samples are directly merged and randomly sampled, the training process can be dominated by a small number of large-scale data sources or frequent attack generators. This may cause the model to overfit specific source distributions and produce clear performance trade-offs across different evaluation sets. Similar data mixing issues have also been widely discussed in multi-domain pretraining and multilingual learning Xie et al. (2023); Chung et al. ; Ye et al. (2025), where the sampling strategy directly affects the generalization ability and robustness of the model. To reduce such data distribution bias, this system adopts attack- and source-balanced data sampling to construct the training distribution of each epoch. For spoofed speech, the system prioritizes the attack generator labels provided by the datasets as sampling units. If such labels are not available, the dataset itself is used as the sampling unit. Let M denote the number of samples for each spoofing unit. In this system, we set M=1000M=1000. If there are K spoofing units in total, the number of spoofed samples in the epoch is MKMK. The system then samples the same number of bonafide speech samples and distributes them uniformly across different real speech datasets. This strategy jointly balances bonafide and spoofed speech, attack generators on the spoofed side, and data sources on the bonafide side. 3.3 Diverse Audio Augmentation In real applications, speech signals often undergo various processing procedures caused by recording devices, room acoustics, background noise, bandwidth limitations, platform compression, and network transmission. These factors may change the observable form of spoofing artifacts and may also make bonafide speech exhibit acoustic conditions different from those in the training set. To improve the robustness of the detector to changes in propagation chains, this system introduces diverse audio augmentation at the waveform level, thereby expanding the coverage of signal conditions during training. Table 2: Audio augmentation operators used during training. For each sample, the system applies augmentation with probability 0.5 by randomly selecting one or two operators. Operator Setting Simulated condition Acoustic- and Recording-related Augmentation RawBoost (Tak et al., 2022a) Algorithms 1, 2, and 3 Device/channel nonlinearity, impulsive disturbance, and colored background noise. RIR (Ko et al., 2017) Room impulse responses Reverberant recordings in enclosed spaces. MUSAN (Snyder et al., 2015) Noise: 5–30 dB; Speech: 15–30 dB Music: 10–30 dB Additive background interference from environmental noise, competing speech, or music. Pitch Shift Pitch offset: [−1,+1][-1,+1] semitones Small global pitch offsets across processing chains. Transmission- and Platform-related Augmentation Filtering Low-pass: 2.2–7.2 kHz; High-pass: 70–360 Hz Band-pass: 300–4000 Hz Narrowband or frequency-shaped transmission paths. Time Masking One waveform segment with 1–10% duration Brief local dropouts or corrupted waveform segments. Codec Compression MP3, Vorbis, Opus, AAC, Speex, GSM, G.711 G.723.1, and Encodec (Défossez et al., 2022) Lossy re-encoding by streaming, messaging, or upload pipelines. Packet Loss Loss rate: 0.01–0.1; Packet length: 20 ms VoIP packet loss with silence or repetition concealment. • Acoustic- and Recording-related Augmentation. The system first considers variations introduced by acoustic environments and recording devices. RawBoost is used to simulate nonlinear channel responses, impulsive perturbations, and colored background noise. Pitch shift introduces slight global pitch changes, RIR convolution introduces room reverberation, and MUSAN mixes noise, speech, or music into the input speech with a random signal-to-noise ratio. These augmentations jointly cover source signals, device responses, spatial acoustics, and external interference, preventing the model from relying only on spoofing cues observed under clean close-talking conditions. • Transmission- and Platform-related Augmentation. The system further considers changes that may occur during speech transmission, uploading, forwarding, and platform processing. Filtering covers low-pass, high-pass, and band-pass conditions, corresponding to transmission paths with bandwidth limitations or frequency-response changes. Time masking simulates local waveform missingness or corruption. Codec compression covers multiple lossy speech or audio coding conditions, while packet loss simulates short packet drops and concealment strategies in VoIP transmission. This class of augmentations is used to improve the robustness of the model to propagation-chain changes and platform-processing distortions. 4 Benchmark Evaluation 4.1 Evaluation Protocol This report uses Speech-DF-Arena Dowerah et al. (2026) as the main evaluation benchmark. The platform aggregates multiple public evaluation sets under a unified evaluation protocol, and we follow its reported results on 14 test sets to evaluate the general detection ability of the model. The evaluation sets cover the ASVspoof series, including ASVspoof19 Wang et al. (2020), ASVspoof21-LA, ASVspoof21-DF Yamagishi et al. (2021), and ASVspoof24-Eval Wang et al. (2024); the ADD challenge series, including ADD 2022 Track 1 (ADD22-T1), ADD 2022 Track 3 (ADD22-T3) Yi et al. (2022), ADD 2023 Round 1 (ADD23-R1), and ADD 2023 Round 2 (ADD23-R2) Yi et al. (2023); as well as In-the-Wild (ITW) Müller et al. (2022), FakeOrReal (FoR) Reimao and Tzerpos (2019), CodecFake (CF) Xie et al. (2025), LibriSeVoc (LSV) Sun et al. (2023), DFADD Du et al. (2024a), and SONAR Li et al. (2024). These test sets span diverse corpus compositions, spoofing generation pipelines, and signal processing conditions. It should be noted that the training corpus used in this report includes the official training partitions of some benchmarks in the evaluation suite. However, many of these benchmarks have clear train–test distribution gaps in spoofing methods, generation models, and other signal-processing conditions. Therefore, the evaluation does not reduce to a purely in-distribution test and still provides evidence of the model’s ability to generalize to unseen attacks. Speech-DF-Arena uses EER, ACC, and F1-score as the main evaluation metrics, and reports dataset-level, pooled, and average results separately. Dataset-level results are computed independently on each test set. Pooled results merge all scores, determine a unified threshold, and compute the metrics accordingly; these results are used as the basis for leaderboard ranking. Average results are obtained by directly averaging the dataset-level results across all test sets. 4.2 Leaderboard Performance Table 3 compares the EER performance of Teffic-Audio with the public systems in the current public snapshot of the official Speech-DF-Arena leaderboard. Complete ACC and F1-score results are provided in Appendix Tables 6 and 7. In terms of pooled EER, Teffic-Audio achieves 1.454%1.454\%, which is lower than all currently public systems. Compared with the top three systems, Modulate-VELMA-2-Synthetic-Voice Modulate (2026) (1.586%1.586\%), Resemble-Detect-3B-Omni Resemble AI (2026b) (2.099%2.099\%), and Hiya-Authenticity-Verification-Multi-v1 Hiya (2026) (2.324%2.324\%), Teffic-Audio reduces the pooled EER by approximately 8.3%8.3\%, 30.7%30.7\%, and 37.4%37.4\% in relative terms, respectively. This result indicates that, under a mainstream detector architecture, the proposed training recipe can provide the detection system with strong cross-domain discriminative ability. From the perspective of individual test sets, Teffic-Audio achieves the lowest EER on five test sets, including CodecFake, ADD 2022 Track 3, ADD 2023 R1, DFADD , and LibriSeVoc, and maintains low error rates on challenging test sets such as ADD 2022 Track 1 and ASV2024-Eval. This shows that the performance of the system is not supported only by advantages on a few test sets, but instead remains stable across multiple types of evaluation conditions. When model size is further considered, Teffic-Audio has 590.0M parameters, which is only larger than Modulate-VELMA-2-Synthetic-Voice and is substantially smaller than Resemble-Detect-3B-Omni (3B) and Hiya-Authenticity-Verification-Multi-v1 (1B). This suggests that the proposed system not only achieves strong spoofing detection performance, but also presents a favorable performance–complexity trade-off. Table 3: EER(%) performance on Speech-DF-Arena. The table reports Teffic-Audio and all systems from the current public leaderboard snapshot, covering licence, parameter size, overall and dataset-level EER results on the 14 evaluation sets. Params denotes the number of parameters in millions. Pooled and Avg. denote pooled EER and the average of dataset-level EERs, respectively. Systems are ranked by pooled EER. ASVspoof ADD System Licence Params Pooled Avg. ITW 19 21-LA 21-DF 24-E FoR CF 22-T1 22-T3 23-R1 23-R2 DFADD LSV SONAR Teffic-Audio Proprietary 590.0 1.454 1.236 1.321 0.242 2.038 0.262 1.405 0.972 0.748 7.031 0.837 0.356 1.836 0.000 0.000 0.251 Modulate-VELMA-2-Synthetic- Voice (Modulate, 2026) Proprietary 316.0 1.586 1.104 1.271 0.299 1.330 0.331 0.384 0.133 1.538 5.059 1.174 1.041 1.742 0.000 0.265 0.888 Resemble-Detect-3B-Omni (Resemble AI, 2026b) Proprietary 3000.0 2.099 2.570 1.347 2.531 3.017 0.579 0.453 2.341 2.689 9.958 1.549 3.390 7.360 0.100 0.000 1.139 Hiya-Authenticity-Verification Multi-v1 (Hiya, 2026) Proprietary 1000.0 2.324 2.113 0.667 0.301 1.006 1.318 0.787 0.000 5.733 12.099 1.188 1.976 4.006 0.017 0.003 0.481 DLMSL-SpeakSure-v0.1 (DLMSL, 2026) Proprietary 658.6 6.142 3.954 1.278 0.042 0.081 0.013 12.894 0.265 6.828 15.140 2.373 5.654 8.280 0.133 0.120 0.074 Whispeak (Whispeak, 2026) Proprietary 98.9 8.060 3.049 1.268 0.394 3.578 3.235 9.924 1.016 0.856 11.942 2.310 2.618 5.007 0.000 0.044 0.533 DF-Raptor (Idiap, 2026) Proprietary 100.0 8.350 8.390 3.191 7.695 12.865 9.288 8.036 2.694 12.731 28.882 3.493 6.452 9.178 1.867 3.442 7.784 DF_Arena_1B_V_1 (Kulkarni et al., 2026) Open 1000.0 9.524 5.919 0.906 1.139 4.657 1.749 17.250 2.915 8.369 22.210 2.204 5.082 11.544 0.000 0.151 1.087 Momenta (Momenta, 2026) Proprietary 350.0 9.763 7.057 4.629 1.208 4.880 1.419 14.346 2.871 8.756 25.471 3.475 10.551 11.098 2.940 1.704 5.445 Syntra Detector (Syntra, 2026) Proprietary 584.0 10.764 6.109 3.978 1.482 14.053 2.026 15.960 0.530 1.187 23.582 2.985 3.653 9.604 0.000 1.171 0.658 DF_Arena_500M_V_1 (Kulkarni et al., 2026) Open 500.0 10.880 5.780 1.760 1.090 4.230 3.300 12.390 2.300 6.360 23.980 2.770 7.470 12.300 0.000 0.120 1.900 MoLEx (A*STAR, 2026) Proprietary 376.4 12.405 9.519 0.034 0.284 6.325 1.885 15.870 0.177 32.401 31.939 3.663 11.111 19.021 6.462 0.303 0.910 Resemble Detect (Resemble AI, 2026a) Proprietary 2112.0 12.747 10.830 3.946 1.321 1.647 3.793 16.291 1.369 33.044 28.218 6.113 21.073 28.278 0.000 1.628 2.989 DF_Arena_100M_V_1 Closed 100.0 13.921 8.398 2.218 1.535 7.612 5.501 21.398 7.420 8.749 27.063 5.421 8.904 17.041 0.000 0.221 4.483 DF_Arena_100M_V_0 Closed 100.0 15.930 10.120 4.290 4.160 10.060 8.830 21.280 6.140 10.390 31.960 6.690 12.690 19.010 0.130 0.420 5.730 XLSR+SLS (Zhang et al., 2024) Open 340.0 16.079 14.015 7.455 0.231 2.869 1.910 18.764 5.080 33.437 33.950 15.744 19.374 21.096 7.542 1.969 24.723 TCM (Truong et al., 2024) Open 319.0 16.691 15.846 7.794 0.188 2.996 2.145 18.850 10.689 36.008 37.403 20.942 23.427 22.743 8.887 2.348 26.572 BiCrossMamba-ST (Kheir et al., 2025) Open 318.2 17.154 15.778 7.937 0.707 3.827 2.347 13.669 6.846 37.703 30.443 18.685 29.444 29.929 8.505 2.120 27.356 Nes2NetX (Liu et al., 2025) Open 317.9 17.366 16.186 7.751 0.122 2.173 1.493 22.060 6.316 39.342 34.471 26.555 21.135 18.450 11.146 2.878 31.536 Wav2Vec2 AASIST (Tak et al., 2022b) Open 317.8 19.607 18.138 11.196 0.221 0.823 6.630 16.247 7.465 43.368 31.050 16.528 27.742 21.929 11.927 11.208 46.124 XLSR Mamba (Tak et al., 2022b) Open 319.0 20.591 14.647 6.709 0.421 0.931 1.884 14.401 6.714 35.266 34.228 19.368 21.848 20.151 10.698 2.234 24.264 Whisper Mesonet (Kawa et al., 2023) Open 7.6 23.551 28.355 26.725 5.833 15.821 2.111 22.549 47.747 34.661 38.383 24.008 41.250 44.562 24.120 15.335 58.663 Wav2Vec2 ECAPA (Kulkarni et al., 2024) Open 324.0 28.995 37.922 30.696 29.695 26.606 22.438 18.656 62.323 40.975 46.435 21.871 35.286 36.702 75.066 28.515 64.566 AASIST (Jung et al., 2022) Open 0.3 32.771 34.404 43.010 0.830 11.461 21.071 35.534 21.643 51.058 47.919 33.187 47.757 32.469 41.860 38.019 57.471 WavLM ECAPA (Kulkarni et al., 2024) Open 102.0 33.463 28.994 34.649 0.761 6.675 15.945 25.992 23.366 46.185 44.169 39.411 29.000 31.937 29.535 31.882 41.944 RawGatST (Tak et al., 2021a) Open 0.4 33.553 34.821 52.538 1.060 10.253 23.262 40.291 53.092 50.000 42.903 32.302 37.866 27.337 23.704 43.364 50.784 RawTFNet (Xiao et al., 2025) Open 0.2 39.938 32.820 38.727 1.890 5.041 16.820 44.740 36.572 51.776 43.850 36.807 38.726 30.539 22.508 29.578 54.786 Hubert ECAPA (Kulkarni et al., 2024) Open 102.0 43.156 33.843 38.658 1.058 12.554 13.793 31.396 33.746 46.224 47.661 39.084 49.569 43.952 34.568 32.033 40.198 Rawnet2 (Tak et al., 2021b) Open 17.6 45.996 47.567 49.187 33.039 40.071 40.669 41.219 65.680 50.219 49.359 47.627 55.641 64.554 39.070 48.167 43.009 5 Ablation Analysis 5.1 Effect of the Training Recipe Table 4 presents the impact of different training strategies on system performance under the fixed architecture. The baseline simply merges multiple speech deepfake datasets and applies random sampling during training, yielding a pooled EER of 7.159% and an average EER of 3.693%. The clear gap between this baseline and the final system indicates that merely increasing the diversity of training data sources is insufficient for robust cross-dataset generalization. After introducing attack- and source-balanced sampling, the pooled EER decreases to 6.456%. This suggests that explicitly balancing attack types and data sources can alleviate distributional bias in multi-source training. With the further inclusion of the supplementary bonafide corpus, the pooled EER is reduced to 5.435%. This result demonstrates that a general-purpose detector requires diverse spoofed samples as well as broad coverage of bonafide speech distributions. For data augmentation, we compare two schemes: using RawBoost alone and using the Diverse Audio Augmentation strategy adopted in this work. Compared with the system after incorporating the supplementary bonafide corpus, using RawBoost alone does not yield consistent or substantial improvements in the overall metrics. A closer examination of individual test sets shows that RawBoost brings clear gains mainly on scenarios such as ASVspoof 2021-LA and ASVspoof 2021-DF. However, it provides limited improvement or even causes degradation on datasets like ASVspoof 2024-Eval. This suggests that RawBoost can simulate certain channel and noise perturbations as a commonly used waveform-level augmentation method in speech deepfake detection. Nevertheless, it is insufficient to consistently improve the cross-dataset generalization of a general-purpose detector under a comprehensive evaluation setting. In contrast, Diverse Audio Augmentation reduces the pooled EER to 1.454% and the average EER to 1.236%, making it a key factor in improving system generalization. The performance gains are particularly pronounced on test sets that are more sensitive to coding compression and channel variations. Specifically, the EER on ASVspoof 2024-Eval decreases from 17.814% to 1.405%, while that on ADD 2022-Track1 decreases from 13.794% to 7.031%. These results indicate that richer audio augmentation can substantially improve the model’s adaptability to complex acoustic conditions. At the same time, no systematic degradation is observed on other test sets. This suggests that the proposed strategy does not weaken the model’s ability to discriminate spoofing cues under conventional speech conditions, thereby leading to stronger cross-dataset robustness. Table 4: Ablation on the training recipe. Architecture is fixed to the w2v-BERT 2.0 encoder with MHASP throughout. Baseline uses merged speech deepfake corpus with random sampling only. ASVspoof ADD Configuration Params Pooled Avg. ITW 19 21-LA 21-DF 24-E FoR CF 22-T1 22-T3 23-R1 23-R2 DFADD LSV SONAR Baseline 590.0 7.159 3.693 1.445 0.600 5.825 1.619 19.650 0.345 0.097 14.885 1.915 1.185 3.655 0.017 0.000 0.459 + Attack- & Source-balanced Sampling 590.0 6.456 3.539 1.173 0.243 6.822 0.540 17.840 0.216 0.144 14.730 1.352 1.502 4.603 0.000 0.000 0.377 + Supplementary Bonafide Corpus 590.0 5.435 3.289 1.242 0.598 7.972 1.304 13.630 0.193 0.235 14.038 1.557 0.581 3.509 0.000 0.000 1.183 + RawBoost Augmentation 590.0 5.896 3.081 1.187 0.178 4.552 0.347 17.814 0.410 0.268 13.794 1.093 0.439 2.046 0.000 0.000 1.006 + Diverse Audio Augmentation 590.0 1.454 1.236 1.321 0.242 2.038 0.262 1.405 0.972 0.748 7.031 0.837 0.356 1.836 0.000 0.000 0.251 5.2 Effect of the Model Architecture Table 5 presents the ablation study on model architecture under our training recipe. We examine three aspects of the detector design: the choice of speech encoder, the utterance-level pooling layer, and the encoder depth of Teffic-Audio. • Encoder Backbone. We first compare the impact of different encoder backbones on system performance. Although traditional convolutional models such as AASIST Jung et al. (2022) and Res2Net Li et al. (2021) have relatively small parameter sizes, they perform substantially worse than detectors based on large-scale SSL-initialized encoders in the comprehensive evaluation setting. Introducing such encoders leads to clear performance improvements. For example, WavLM Large Chen et al. (2022) + MHASP and XLS-R 1B Xiao and Das (2025) + MHASP achieve pooled EERs of 4.332% and 4.419%, respectively, outperforming AASIST and Res2Net. A further comparison between different scales of the same encoder family shows that WavLM Base + MHASP and XLS-R 300M + MHASP obtain pooled EERs of 9.446% and 6.079%, respectively, both significantly higher than their corresponding large-scale variants. This indicates that increasing the encoder scale generally helps improve system performance. Among different SSL-initialized backbones, the w2v-BERT 2.0 encoder adopted in Teffic-Audio achieves the best result, further reducing the pooled EER to 1.454%. It also shows stronger generalization across multiple test sets. These results suggest that the performance of general-purpose speech deepfake detection is not determined solely by parameter size. It is also closely related to the encoder’s pretraining objective and acoustic representation capability. From another perspective, under the same evaluation protocol, models trained with our recipe also substantially outperform leaderboard systems that use similar backbones. Specifically, AASIST achieves a pooled EER of 15.668%, corresponding to a 52.2% reduction compared with the leaderboard AASIST system. WavLM Base + MHASP achieves 9.446%, reducing the EER by 71.7% compared with WavLM-ECAPA (Kulkarni et al., 2024). XLS-R 300M + MHASP achieves 6.079%, reducing the EER by 62.2% compared with XLSR+SLS (Zhang et al., 2024). These results further indicate that the proposed training recipe can effectively improve the overall performance of different encoder backbones in general-purpose deepfake detection. • Pooling Layer. We next compare the impact of different utterance-level pooling layers. With the w2v-BERT 2.0 encoder fixed, mean pooling and ASP achieve pooled EERs of 2.269% and 2.339%, respectively. Both yield higher EERs than the final MHASP-based system, indicating that the choice of pooling layer also affects overall system performance. Notably, although ASP introduces attention-based pooling, it does not improve over mean pooling. This suggests that simple frame-level weighting may be insufficient to exploit the fine-grained representations produced by the encoder. In contrast, MHASP aggregates frame-level statistics from different subspaces through multiple attention heads, which provides a more effective utterance-level representation for cross-dataset generalization. • Encoder Depth. The preceding backbone comparison shows that directly adopting smaller SSL backbones leads to clear performance degradation. We therefore further examine whether the Teffic-Audio encoder can be made shallower to reduce parameter size while maintaining stable cross-dataset generalization. To this end, we train and evaluate variants with different numbers of Conformer blocks. The results show that overall performance improves as encoder depth increases. From N=3N=3 to the full 24-layer configuration, the pooled EER gradually decreases from 4.735% to 1.454%. This result indicates that, after selecting an effective backbone, encoder depth remains a key factor in the trade-off between performance and complexity. Further examination on individual test sets shows that the benefits of increasing depth are not uniform across all evaluation scenarios. On test sets such as ITW, ASVspoof21-DF, DFADD, and LSV, the 4-layer or 6-layer configurations already achieve relatively low EERs. This suggests that spoofing cues in many milder test conditions can already be captured by a shallower encoder. In contrast, the advantage of the full-depth model is mainly observed on more challenging test sets, such as ASVspoof24-E and ADD22-T1. For example, the EER on ASVspoof24-E decreases from 6.744% with N=4N=4 to 1.405% with the full model. The EER on ADD22-T1 decreases from 10.734% to 7.031%. These results suggest that increasing encoder depth mainly improves the model’s adaptability to challenging scenarios. Meanwhile, a clear degradation is observed when reducing the depth from N=4N=4 to N=3N=3. The pooled EER increases from 3.346% to 4.735%, corresponding to a relative increase of 41.5%. This indicates that overly reducing the encoder depth weakens the stability of cross-dataset discrimination. It is worth noting that shallower Teffic-Audio configurations still retain strong performance. The 4-layer configuration contains only 106.4M parameters, yet achieves a pooled EER of 3.346%. This outperforms WavLM Base Chen et al. (2022) + MHASP, which has a comparable parameter size. The 6-layer configuration further achieves a pooled EER of 2.839% with 154.8M parameters. This performance is close to leading systems on the leaderboard, such as Resemble-Detect-3B-Omni and Hiya-Authenticity-Verification-Multi-v1. These results indicate that reducing the depth of an effective backbone provides a better performance–complexity trade-off than directly replacing it with a smaller SSL encoder. Combined with an appropriate training recipe, shallower configurations can still maintain strong cross-dataset generalization. Table 5: Ablation study on model architecture. Teffic-Audio refers to the full system using MHASP and a 24-layer encoder initialized from w2v-BERT 2.0. All metric columns report EER (%). MP and ASP denote mean pooling and attentive statistics pooling, respectively. For Teffic-Audio depth variants, N denotes the encoder depth, and the N-layer encoders are initialized from the first N Conformer blocks of w2v-BERT 2.0. ASVspoof ADD Configuration Params Pooled Avg. ITW 19 21-LA 21-DF 24-E FoR CF 22-T1 22-T3 23-R1 23-R2 DFADD LSV SONAR Teffic-Audio 590.0 1.454 1.236 1.321 0.242 2.038 0.262 1.405 0.972 0.748 7.031 0.837 0.356 1.836 0.000 0.000 0.251 Encoder Backbone AASIST 0.3 15.668 11.480 2.225 16.228 20.644 11.346 15.896 9.189 13.504 29.852 6.730 9.827 16.002 0.664 1.325 7.294 Res2Net + MHASP 2.8 15.023 10.744 8.847 4.639 21.298 23.619 22.809 11.734 0.864 27.747 4.952 5.109 8.341 0.000 0.189 10.261 WavLM Base + MHASP 117.7 9.446 8.122 3.449 6.254 12.162 4.957 4.530 6.042 10.854 30.250 5.275 8.514 18.639 0.133 0.032 2.611 WavLM Large + MHASP 315.4 4.332 4.005 1.700 2.083 7.570 2.246 4.764 4.489 1.903 14.158 2.143 5.503 8.274 0.000 0.000 1.235 XLS-R 300M + MHASP 339.8 6.079 4.519 2.074 2.294 8.672 5.104 5.101 4.142 2.836 15.581 1.985 5.010 9.102 0.000 0.006 1.361 XLS-R 1B + MHASP 962.5 4.419 2.909 1.375 0.493 2.249 1.129 7.877 1.552 2.164 14.877 2.327 1.413 4.909 0.000 0.000 0.355 Pooling Layer Teffic-Audio w/ MP 587.9 2.269 1.515 1.234 0.244 2.666 0.255 4.727 0.474 0.898 6.821 0.769 0.436 2.057 0.000 0.000 0.459 Teffic-Audio w/ ASP 590.5 2.339 1.499 1.036 0.247 3.036 0.264 4.514 0.798 0.710 6.803 0.712 0.443 1.946 0.000 0.000 0.481 Encoder Depth Teffic-Audio (N=3N=3) 82.2 4.735 3.280 1.459 2.423 6.449 1.835 9.901 2.785 0.594 13.075 1.340 0.798 3.683 0.000 0.000 1.582 Teffic-Audio (N=4N=4) 106.4 3.346 2.415 1.083 0.699 4.244 0.593 6.744 4.726 0.595 10.734 1.059 0.440 2.230 0.000 0.000 0.658 Teffic-Audio (N=6N=6) 154.8 2.839 2.005 0.922 0.515 3.134 0.484 5.306 2.633 0.937 9.640 0.918 0.526 2.153 0.000 0.000 0.910 Teffic-Audio (N=12N=12) 299.9 2.317 1.965 1.030 0.565 3.307 0.455 3.769 4.057 1.279 8.754 0.826 0.576 2.324 0.000 0.000 0.564 6 Conclusion This report presented Teffic-Audio, a general speech deepfake detection system for comprehensive evaluation environment. The system follows a standard detection pipeline and focuses on the construction of an effective training distribution. Trained only with open-source data, Teffic-Audio achieves a pooled EER of 1.454% on the 14 test sets of Speech-DF-Arena and ranks first compared with all currently public systems. Ablation studies show that the final performance cannot be attributed merely to large-scale data aggregation. Balanced sampling, supplementary bonafide speech, and especially diverse audio augmentation all contribute to stronger cross-dataset generalization. The architecture analysis further shows that the encoder backbone and pooling layer have clear effects on system performance. In particular, reducing the depth of the system backbone still yields competitive results while substantially decreasing the number of parameters. These findings show that Teffic-Audio provides an effective and practical baseline for general speech deepfake detection under comprehensive evaluation environment. Authors Contributors: Wan Lin, Li Wang, Jindong Wang, Kunyu Feng, Zhizheng Wu. References A*STAR (2026) MoLEx. Note: https://w.a-star.edu.sg/Speech deepfake detection system. Accessed May 11, 2026 Cited by: Table 6, Table 7, Table 3. R. Ardila, M. Branson, K. Davis, M. Kohler, J. Meyer, M. Henretty, R. Morais, L. Saunders, F. Tyers, and G. Weber (2020) Common voice: a massively-multilingual speech corpus. In Proceedings of the twelfth language resources and evaluation conference, p. 4218–4222. Cited by: Table 1. A. Babu, C. Wang, A. Tjandra, K. Lakhotia, Q. Xu, N. Goyal, K. Singh, P. Von Platen, Y. Saraf, J. Pino, et al. (2021) XLS-r: self-supervised cross-lingual speech representation learning at scale. arXiv preprint arXiv:2111.09296. Cited by: §1. A. Baevski, Y. Zhou, A. Mohamed, and M. Auli (2020) Wav2vec 2.0: a framework for self-supervised learning of speech representations. Advances in neural information processing systems 33, p. 12449–12460. Cited by: §1. L. Barrault, Y. Chung, M. C. Meglioli, D. Dale, N. Dong, M. Duppenthaler, P. Duquenne, B. Ellis, H. Elsahar, J. Haaheim, et al. (2023) Seamless: multilingual expressive and streaming speech translation. arXiv preprint arXiv:2312.05187. Cited by: §2. N. A. Chandra, R. Murtfeldt, L. Qiu, A. Karmakar, H. Lee, E. Tanumihardja, K. Farhat, B. Caffee, S. Paik, C. Lee, et al. (2025) Deepfake-eval-2024: a multi-modal in-the-wild benchmark of deepfakes circulated in 2024. arXiv preprint arXiv:2503.02857. Cited by: §1. G. Chen, S. Chai, G. Wang, J. Du, W. Zhang, C. Weng, D. Su, D. Povey, J. Trmal, J. Zhang, et al. (2021) Gigaspeech: an evolving, multi-domain asr corpus with 10,000 hours of transcribed audio. arXiv preprint arXiv:2106.06909. Cited by: Table 1. S. Chen, S. Liu, L. Zhou, Y. Liu, X. Tan, J. Li, S. Zhao, Y. Qian, and F. Wei (2024) Vall-e 2: neural codec language models are human parity zero-shot text to speech synthesizers. arXiv preprint arXiv:2406.05370. Cited by: §1. S. Chen, C. Wang, Z. Chen, Y. Wu, S. Liu, Z. Chen, J. Li, N. Kanda, T. Yoshioka, X. Xiao, et al. (2022) Wavlm: large-scale self-supervised pre-training for full stack speech processing. IEEE Journal of Selected Topics in Signal Processing 16 (6), p. 1505–1518. Cited by: §1, 1st item, 3rd item. Y. Chen, Z. Niu, Z. Ma, K. Deng, C. Wang, J. JianZhao, K. Yu, and X. Chen (2025) F5-tts: a fairytaler that fakes fluent and faithful speech with flow matching. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), p. 6255–6271. Cited by: §1. [11] H. W. Chung, X. Garcia, A. Roberts, Y. Tay, O. Firat, S. Narang, and N. Constant UniMax: fairer and more effective language sampling for large-scale multilingual pretraining. In The Eleventh International Conference on Learning Representations, Cited by: §3.2. A. Cohen, I. Rimon, E. Aflalo, and H. H. Permuter (2022) A study on data augmentation in voice anti-spoofing. Speech Communication 141, p. 56–67. Cited by: §1. A. Défossez, J. Copet, G. Synnaeve, and Y. Adi (2022) High fidelity neural audio compression. arXiv preprint arXiv:2210.13438. Cited by: Table 2. DLMSL (2026) DLMSL-SpeakSure-v0.1. Note: https://dlmsl.csie.ncu.edu.tw/Commercial or online speech authenticity verification system. Accessed May 11, 2026 Cited by: Table 6, Table 7, Table 3. S. Dowerah, A. Kulkarni, A. Kulkarni, H. M. Tran, J. Kalda, A. Fedorchenko, B. Fauve, D. Lolive, T. Alumäe, and M. M. Doss (2026) Speech df arena: a leaderboard for speech deepfake detection models. IEEE Open Journal of Signal Processing. Cited by: §1, §1, §4.1. J. Du, I. Lin, I. Chiu, X. Chen, H. Wu, W. Ren, Y. Tsao, H. Lee, and J. R. Jang (2024a) Dfadd: the diffusion and flow-matching based audio deepfake dataset. In 2024 IEEE Spoken Language Technology Workshop (SLT), p. 921–928. Cited by: Table 1, §4.1. Z. Du, Y. Wang, Q. Chen, X. Shi, X. Lv, T. Zhao, Z. Gao, Y. Yang, C. Gao, H. Wang, et al. (2024b) Cosyvoice 2: scalable streaming speech synthesis with large language models. arXiv preprint arXiv:2412.10117. Cited by: §1. Y. Fan, J. Kang, L. Li, K. Li, H. Chen, S. Cheng, P. Zhang, Z. Zhou, Y. Cai, and D. Wang (2020) Cn-celeb: a challenging chinese speaker recognition dataset. In ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), p. 7604–7608. Cited by: Table 1. J. Frank and L. Schönherr (2021) Wavefake: a data set to facilitate audio deepfake detection. arXiv preprint arXiv:2111.02813. Cited by: §1, Table 1. W. Ge, X. Wang, X. Liu, and J. Yamagishi (2025) Post-training for deepfake speech detection. arXiv preprint arXiv:2506.21090. Cited by: §1. A. Gulati, J. Qin, C. Chiu, N. Parmar, Y. Zhang, J. Yu, W. Han, S. Wang, Z. Zhang, Y. Wu, et al. (2020) Conformer: convolution-augmented transformer for speech recognition. In Proc. Interspeech 2020, p. 5036–5040. Cited by: §1, §2. Hiya (2026) Hiya-Authenticity-Verification-Multi-v1. Note: https://w.hiya.com/Commercial authenticity verification model. Accessed May 11, 2026 Cited by: Table 6, Table 7, §4.2, Table 3. P. Huang, S. Wang, S. Yao, and M. Zhou (2026) Generalizable speech deepfake detection via information bottleneck enhanced adversarial alignment. In ICASSP 2026-2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), p. 19087–19091. Cited by: §1. W. Huang, Y. Gu, Z. Wang, H. Zhu, and Y. Qian (2025a) Generalizable audio deepfake detection via latent space refinement and augmentation. In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), p. 1–5. Cited by: §1. W. Huang, Y. Gu, Z. Wang, H. Zhu, and Y. Qian (2025b) SpeechFake: a large-scale multilingual speech deepfake dataset incorporating cutting-edge generation methods. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), p. 9985–9998. Cited by: Table 1. W. Huang, Y. Mao, and Y. Qian (2025c) A data-centric approach to generalizable speech deepfake detection. arXiv preprint arXiv:2512.18210. Cited by: §1. Idiap (2026) DF-Raptor. Note: https://w.idiap.ch/en/Speech deepfake detection system. Accessed May 11, 2026 Cited by: Table 6, Table 7, Table 3. Z. Ju, Y. Wang, K. Shen, X. Tan, D. Xin, D. Yang, E. Liu, Y. Leng, K. Song, S. Tang, et al. (2024) NaturalSpeech 3: zero-shot speech synthesis with factorized codec and diffusion models. In International Conference on Machine Learning, p. 22605–22623. Cited by: §1. J. Jung, H. Heo, H. Tak, H. Shim, J. S. Chung, B. Lee, H. Yu, and N. Evans (2022) Aasist: audio anti-spoofing using integrated spectro-temporal graph attention networks. In ICASSP 2022-2022 IEEE international conference on acoustics, speech and signal processing (ICASSP), p. 6367–6371. Cited by: Table 6, Table 7, §1, Table 3, 1st item. J. Jung, Y. Wu, X. Wang, J. Kim, S. Maiti, Y. Matsunaga, H. Shim, J. Tian, N. Evans, J. S. Chung, et al. (2025) Spoofceleb: speech deepfake detection and sasv in the wild. IEEE Open Journal of Signal Processing. Cited by: Table 1. P. Kawa, M. Plata, M. Czuba, P. Syga, et al. (2023) Improved deepfake detection using whisper features. arXiv preprint arXiv:2306.01428. Cited by: Table 6, Table 7, Table 3. Y. E. Kheir, T. Polzehl, and S. Möller (2025) BiCrossMamba-st: speech deepfake detection with bidirectional mamba spectro-temporal cross-attention. arXiv preprint arXiv:2505.13930. Cited by: Table 6, Table 7, Table 3. T. Ko, V. Peddinti, D. Povey, M. L. Seltzer, and S. Khudanpur (2017) A study on data augmentation of reverberant speech for robust speech recognition. In 2017 IEEE international conference on acoustics, speech and signal processing (ICASSP), p. 5220–5224. Cited by: Table 2. A. Kulkarni, S. Dowerah, A. Kulkarni, T. Alumäe, and M. M. Doss (2026) Do compact ssl backbones matter for audio deepfake detection? a controlled study with raptor. External Links: 2603.06164 Cited by: Table 6, Table 6, Table 7, Table 7, §1, Table 3, Table 3. A. Kulkarni, H. M. Tran, A. Kulkarni, S. Dowerah, D. Lolive, and M. M. Doss (2024) Exploring generalization to unseen audio data for spoofing: insights from ssl models. In ASVSpoof workshop 2024, Cited by: Table 6, Table 6, Table 6, Table 7, Table 7, Table 7, Table 3, Table 3, Table 3, 1st item. C. Y. Kwok, J. Q. Yip, Z. Qiu, C. H. Chi, and K. Y. Lam (2025) Bona fide cross testing reveals weak spot in audio deepfake detection systems. arXiv preprint arXiv:2509.09204. Cited by: 2nd item. J. Laakkonen, I. Kukanov, and V. Hautamäki (2026) Generalizable speech deepfake detection via meta-learned lora. In ICASSP 2026-2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), p. 19092–19096. Cited by: §1. M. Li, Y. Ahmadiadli, and X. Zhang (2025a) A survey on speech deepfake detection. ACM Computing Surveys 57 (7), p. 1–38. Cited by: §1. M. Li, X. Zhang, and L. Zhao (2025b) Frame-level temporal difference learning for partial deepfake speech detection. IEEE Signal Processing Letters. Cited by: §1. X. Li, P. Chen, and W. Wei (2024) Sonar: a synthetic ai-audio detection framework and benchmark. Cited by: §4.1. X. Li, N. Li, C. Weng, X. Liu, D. Su, D. Yu, and H. Meng (2021) Replay and synthetic speech detection with res2net architecture. In ICASSP 2021-2021 IEEE international conference on acoustics, speech and signal processing (ICASSP), p. 6354–6358. Cited by: §1, 1st item. T. Liu, D. Truong, R. K. Das, K. A. Lee, and H. Li (2025) Nes2net: a lightweight nested architecture for foundation model driven speech anti-spoofing. IEEE Transactions on Information Forensics and Security 20, p. 12005–12018. Cited by: Table 6, Table 7, Table 3. Modulate (2026) Modulate-VELMA-2-Synthetic-Voice. Note: https://w.modulate.ai/benchmarksSpeech DF Arena leaderboard entry, added March 11, 2026. Accessed May 11, 2026 Cited by: Table 6, Table 7, §4.2, Table 3. Momenta (2026) Momenta. Note: https://momenta.network/Speech deepfake detection system. Accessed May 11, 2026 Cited by: Table 6, Table 7, Table 3. N. Müller, P. Kawa, W. Choong, A. Stan, A. T. Bukkapatnam, K. Pizzi, A. Wagner, and P. Sperl (2025) Replay attacks against audio deepfake detection. In Proc. Interspeech 2025, p. 2245–2249. Cited by: Table 1. N. M. Müller, P. Czempin, F. Dieckmann, A. Froghyar, and K. Böttinger (2022) Does audio deepfake detection generalize?. arXiv preprint arXiv:2203.16263. Cited by: §1, §4.1. N. M. Müller, N. Evans, H. Tak, P. Sperl, and K. Böttinger (2024a) Harder or different? understanding generalization of audio deepfake detection. In Proc. Interspeech 2024, p. 2705–2709. Cited by: §1. N. M. Müller, P. Kawa, W. H. Choong, E. Casanova, E. Gölge, T. Müller, P. Syga, P. Sperl, and K. Böttinger (2024b) Mlaad: the multi-language audio anti-spoofing dataset. In 2024 International Joint Conference on Neural Networks (IJCNN), p. 1–7. Cited by: §1, Table 1. V. Negroni, D. Salvi, A. I. Mezza, P. Bestagini, and S. Tubaro (2025) Leveraging mixture of experts for improved speech deepfake detection. In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), p. 1–5. Cited by: §1. K. Okabe, T. Koshinaka, and K. Shinoda (2018) Attentive statistics pooling for deep speaker embedding. arXiv preprint arXiv:1803.10963. Cited by: §1, §2. Z. Pan, S. H. Bhupendra, and J. Wu (2025) MoLEx: mixture of lora experts in speech self-supervised models for audio deepfake detection. arXiv preprint arXiv:2509.09175. Cited by: §1. V. Panayotov, G. Chen, D. Povey, and S. Khudanpur (2015) Librispeech: an asr corpus based on public domain audio books. In 2015 IEEE international conference on acoustics, speech and signal processing (ICASSP), p. 5206–5210. Cited by: Table 1. R. Reimao and V. Tzerpos (2019) FoR: a dataset for synthetic speech detection. In 2019 International Conference on Speech Technology and Human-Computer Dialogue (SpeD), p. 1–10. Cited by: Table 1, §4.1. Resemble AI (2026a) Resemble Detect. Note: https://w.resemble.ai/Synthetic speech detection system. Accessed May 11, 2026 Cited by: Table 6, Table 7, Table 3. Resemble AI (2026b) Resemble-Detect-3B-Omni. Note: https://w.resemble.ai/Commercial synthetic speech detection model. Accessed May 11, 2026 Cited by: Table 6, Table 7, §4.2, Table 3. H. Shi, X. Shi, S. Dogan, S. Alzubi, T. Huang, and Y. Zhang (2025) Benchmarking audio deepfake detection robustness in real-world communication scenarios. In 2025 33rd European Signal Processing Conference (EUSIPCO), p. 566–570. Cited by: §1. Y. Shi, H. Bu, X. Xu, S. Zhang, and M. Li (2020) Aishell-3: a multi-speaker mandarin tts corpus and the baselines. arXiv preprint arXiv:2010.11567. Cited by: Table 1. D. Snyder, G. Chen, and D. Povey (2015) Musan: a music, speech, and noise corpus. arXiv preprint arXiv:1510.08484. Cited by: Table 2. C. Sun, S. Jia, S. Hou, and S. Lyu (2023) Ai-synthesized voice detection using neural vocoder artifacts. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, p. 904–912. Cited by: §1, Table 1, §4.1. Syntra (2026) Syntra Detector. Note: https://syntra.io/Synthetic speech detection system. Accessed May 11, 2026 Cited by: Table 6, Table 7, Table 3. H. Tak, J. Jung, J. Patino, M. Kamble, M. Todisco, and N. Evans (2021a) End-to-end spectro-temporal graph attention networks for speaker verification anti-spoofing and speech deepfake detection. arXiv preprint arXiv:2107.12710. Cited by: Table 6, Table 7, Table 3. H. Tak, M. Kamble, J. Patino, M. Todisco, and N. Evans (2022a) Rawboost: a raw data boosting and augmentation method applied to automatic speaker verification anti-spoofing. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), p. 6382–6386. Cited by: §1, Table 2. H. Tak, J. Patino, M. Todisco, A. Nautsch, N. Evans, and A. Larcher (2021b) End-to-end anti-spoofing with rawnet2. In ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), p. 6369–6373. Cited by: Table 6, Table 7, §1, Table 3. H. Tak, M. Todisco, X. Wang, J. Jung, J. Yamagishi, and N. Evans (2022b) Automatic speaker verification spoofing and deepfake detection using wav2vec 2.0 and data augmentation. arXiv preprint arXiv:2202.12233. Cited by: Table 6, Table 6, Table 7, Table 7, Table 3, Table 3. H. M. Tran, D. Lolive, D. Guennec, A. Sini, A. Delhay, and P. Marteau (2025) Leveraging ssl speech features and mamba for enhanced deepfake detection. In Interspeech 2025, p. 5323–5327. Cited by: §1. D. Truong, R. Tao, T. Nguyen, H. Luong, K. A. Lee, and E. S. Chng (2024) Temporal-channel modeling in multi-head self-attention for synthetic speech detection. arXiv preprint arXiv:2406.17376. Cited by: Table 6, Table 7, §1, Table 3. X. Wang, H. Delgado, H. Tak, J. Jung, H. Shim, M. Todisco, I. Kukanov, X. Liu, M. Sahidullah, T. H. Kinnunen, et al. (2024) ASVspoof 5: crowdsourced speech data, deepfakes, and adversarial attacks at scale. In Proc. ASVspoof 2024, p. 1–8. Cited by: §1, Table 1, §4.1. X. Wang, J. Yamagishi, M. Todisco, H. Delgado, A. Nautsch, N. Evans, M. Sahidullah, V. Vestman, T. Kinnunen, K. A. Lee, et al. (2020) ASVspoof 2019: a large-scale public database of synthesized, converted and replayed speech. Computer Speech & Language 64, p. 101114. External Links: Document Cited by: §1, §1, Table 1, §4.1. Z. Wang, R. Fu, Z. Wen, J. Tao, X. Wang, Y. Xie, X. Qi, S. Shi, Y. Lu, Y. Liu, et al. (2025) Mixture of experts fusion for fake audio detection using frozen wav2vec 2.0. In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), p. 1–5. Cited by: §1. Whispeak (2026) Whispeak. Note: https://whispeak.io/Voice authentication and speech security system. Accessed May 11, 2026 Cited by: Table 6, Table 7, Table 3. Z. Wu, R. K. Das, J. Yang, and H. Li (2020) Light convolutional neural network with feature genuinization for detection of synthetic speech attacks. In Proc. Interspeech 2020, p. 1101–1105. Cited by: §1. Z. Wu, T. Kinnunen, N. Evans, J. Yamagishi, C. Hanilçi, and M. Sahidullah (2015) ASVspoof 2015: the first automatic speaker verification spoofing and countermeasures challenge. p. . External Links: Document Cited by: §1, §1, Table 1. Y. Xiao, T. Dang, and R. K. Das (2025) RawTFNet: a lightweight cnn architecture for speech anti-spoofing. In 2025 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), p. 1997–2001. Cited by: Table 6, Table 7, Table 3. Y. Xiao and R. K. Das (2025) XLSR-mamba: a dual-column bidirectional state space model for spoofing attack detection. IEEE Signal Processing Letters. Cited by: 1st item. S. M. Xie, H. Pham, X. Dong, N. Du, H. Liu, Y. Lu, P. S. Liang, Q. V. Le, T. Ma, and A. W. Yu (2023) Doremi: optimizing data mixtures speeds up language model pretraining. Advances in Neural Information Processing Systems 36, p. 69798–69818. Cited by: §3.2. Y. Xie, Y. Lu, R. Fu, Z. Wen, Z. Wang, J. Tao, X. Qi, X. Wang, Y. Liu, H. Cheng, et al. (2025) The codecfake dataset and countermeasures for the universally detection of deepfake audio. IEEE Transactions on Audio, Speech and Language Processing 33, p. 386–400. Cited by: §1, Table 1, §4.1. J. Yamagishi, X. Wang, M. Todisco, M. Sahidullah, J. Patino, A. Nautsch, X. Liu, K. A. Lee, T. Kinnunen, N. Evans, et al. (2021) ASVspoof 2021: accelerating progress in spoofed and deepfake speech detection. In Proc. ASVSPOOF 2021, p. 47–54. Cited by: §1, §4.1. M. Yang, Y. Gu, Q. He, Y. Li, P. Zhang, Y. Chen, Z. Wang, H. Zhu, J. Liu, and W. Wang (2025) Generalizable audio deepfake detection via hierarchical structure learning and feature whitening in poincaré sphere. In Proc. Interspeech 2025, p. 2255–2259. Cited by: §1. J. Ye, P. Liu, T. Sun, J. Zhan, Y. Zhou, and X. Qiu (2025) Data mixing laws: optimizing data mixtures by predicting language modeling performance. In International Conference on Learning Representations, Vol. 2025, p. 82263–82287. Cited by: §3.2. J. Yi, R. Fu, J. Tao, S. Nie, H. Ma, C. Wang, T. Wang, Z. Tian, Y. Bai, C. Fan, et al. (2022) Add 2022: the first audio deep synthesis detection challenge. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), p. 9216–9220. Cited by: §1, Table 1, §4.1. J. Yi, J. Tao, R. Fu, X. Yan, C. Wang, T. Wang, C. Y. Zhang, X. Zhang, Y. Zhao, Y. Ren, et al. (2023) Add 2023: the second audio deepfake detection challenge. arXiv preprint arXiv:2305.13774. Cited by: §1, Table 1, §4.1. Q. Zhang, S. Wen, and T. Hu (2024) Audio deepfake detection with self-supervised xls-r and sls classifier. In Proceedings of the 32nd ACM International Conference on Multimedia, p. 6765–6773. Cited by: Table 6, Table 7, §1, Table 3, 1st item. Appendix A Additional Leaderboard Performances Tables 6 and 7 report additional ACC and F1-score comparisons on the Speech-DF-Arena. Teffic-Audio achieves the best pooled ACC and pooled F1-score among the listed systems. Together with the EER results in the main text, these results show a consistent performance advantage across different measurements, further supporting the stability of Teffic-Audio in the comprehensive cross-dataset evaluation. Table 6: ACC performance on Speech-DF-Arena. ASVspoof ADD System Licence Params Pooled Avg. ITW 19 21-LA 21-DF 24-E FoR CF 22-T1 22-T3 23-R1 23-R2 DFADD LSV SONAR Teffic-Audio Proprietary 590.0 98.546 98.759 98.682 99.761 97.961 99.738 98.595 99.006 99.251 92.969 99.162 99.645 98.163 99.973 99.995 99.721 Modulate-VELMA-2-Synthetic- Voice [Modulate, 2026] Proprietary 316.0 98.414 98.897 98.726 99.702 98.670 99.667 99.616 99.890 98.463 94.940 98.825 98.958 98.257 99.973 99.730 99.139 Resemble-Detect-3B-Omni [Resemble AI, 2026b] Proprietary 3000.0 97.901 97.430 98.656 97.469 96.984 99.421 99.547 97.637 97.312 90.043 98.453 96.611 92.641 99.947 99.995 98.835 Hiya-Authenticity-Verification Multi-v1 [Hiya, 2026] Proprietary 1000.0 97.680 97.880 99.330 99.700 98.990 98.680 99.210 99.800 94.270 87.900 98.810 98.020 96.000 99.950 99.990 99.540 DLMSL-SpeakSure-v0.1 [DLMSL, 2026] Proprietary 658.6 93.858 96.043 98.726 99.959 99.920 99.986 87.105 99.757 93.173 84.859 97.627 94.347 91.721 99.840 99.870 99.899 DF-Raptor [Idiap, 2026] Proprietary 100.0 92.330 92.080 96.523 95.870 93.890 98.025 93.140 96.952 84.447 69.436 96.215 92.974 93.335 97.870 92.311 88.222 Whispeak [Whispeak, 2026] Proprietary 98.9 91.953 96.947 98.729 99.604 96.420 96.766 90.075 99.006 99.144 88.057 97.689 97.381 94.994 99.973 99.946 99.443 DF_Arena_1B_V_1 [Kulkarni et al., 2026] Open 1000.0 90.476 94.079 99.097 98.864 95.342 98.251 82.750 97.107 91.631 77.791 97.798 94.917 88.457 99.973 99.843 98.886 Momenta [Momenta, 2026] Proprietary 350.0 90.236 92.947 95.374 98.794 95.119 98.581 85.654 97.151 91.244 74.530 96.524 89.450 88.901 97.071 98.291 94.580 Syntra Detector [Syntra, 2026] Proprietary 584.0 89.236 93.893 96.019 98.516 85.946 97.972 84.040 99.492 98.816 76.417 97.013 96.348 90.397 99.973 98.837 99.367 DF_Arena_500M_V_1 [Kulkarni et al., 2026] Open 500.0 89.110 94.190 98.240 98.910 95.770 96.700 87.610 97.730 93.650 76.020 97.230 92.530 87.700 99.970 99.870 98.130 MoLEx [A*STAR, 2026] Proprietary 376.4 87.595 90.483 99.969 99.719 93.673 98.114 84.130 99.845 67.600 68.062 96.338 88.890 80.978 93.582 99.692 99.063 Resemble Detect [Resemble AI, 2026a] Proprietary 2112.0 87.253 89.170 96.051 98.679 98.353 96.207 83.709 98.653 66.959 71.781 93.886 78.928 71.723 99.973 98.366 97.036 DF_Arena_100M_V_1 Closed 100.0 86.079 91.605 97.778 98.467 92.389 94.499 78.601 92.602 91.252 72.937 94.578 91.097 82.960 99.973 99.789 95.542 DF_Arena_100M_V_0 Closed 100.0 84.060 89.070 95.710 95.830 89.940 91.170 78.720 93.840 89.610 68.050 93.310 87.310 80.990 99.890 99.580 94.300 XLSR+SLS [Zhang et al., 2024] Open 340.0 83.921 85.984 92.542 99.770 97.129 98.089 81.236 94.898 66.563 66.051 84.258 80.624 78.905 92.437 98.026 75.304 TCM [Truong et al., 2024] Open 319.0 83.309 84.153 92.209 99.815 97.006 97.854 81.150 89.289 63.992 62.599 79.060 76.574 77.256 91.079 97.647 73.455 BiCrossMamba-ST [Kheir et al., 2025] Open 318.2 82.846 84.219 92.061 99.291 96.172 97.653 86.331 93.132 62.298 69.556 81.316 70.557 70.070 91.505 97.874 72.619 Nes2NetX [Liu et al., 2025] Open 317.9 82.634 83.814 92.253 99.879 97.828 98.507 77.940 93.706 60.659 65.530 73.443 78.864 81.549 88.815 97.117 68.490 Wav2Vec2 AASIST [Tak et al., 2022b] Open 317.8 80.393 81.867 88.807 99.775 99.176 93.371 83.753 92.557 56.634 68.949 83.472 72.257 78.072 88.096 88.787 53.901 XLSR Mamba [Tak et al., 2022b] Open 319.0 79.409 85.355 93.294 99.577 99.068 98.115 85.599 93.308 64.736 65.771 80.631 78.152 79.850 89.348 97.761 75.709 Whisper Mesonet [Kawa et al., 2023] Open 7.6 76.449 71.649 73.272 94.166 84.178 97.889 77.451 52.275 65.338 61.616 75.991 58.751 55.437 75.899 84.659 41.363 Wav2Vec2 ECAPA [Kulkarni et al., 2024] Open 324.0 71.005 62.084 69.307 70.303 73.395 77.561 81.344 37.699 59.026 53.566 78.128 64.713 63.299 24.980 71.477 35.461 AASIST [Jung et al., 2022] Open 0.3 67.229 65.597 56.994 99.169 88.539 78.930 64.466 78.379 48.941 52.080 66.813 52.244 67.530 58.109 61.973 42.553 WavLM ECAPA [Kulkarni et al., 2024] Open 102.0 66.536 71.010 65.348 99.238 93.324 84.056 74.008 76.656 53.816 55.830 60.588 71.001 68.064 70.493 68.113 58.080 RawGatST [Tak et al., 2021a] Open 0.4 66.447 65.181 47.465 98.939 89.747 76.740 59.709 46.930 50.002 57.098 67.698 62.135 72.662 76.272 56.624 49.240 RawTFNet [Xiao et al., 2025] Open 0.2 60.062 67.178 61.276 98.109 94.961 83.180 55.260 63.450 48.223 56.151 63.192 61.275 69.460 77.470 70.412 45.187 Hubert ECAPA [Kulkarni et al., 2024] Open 102.0 56.844 66.155 61.339 98.946 87.445 86.208 68.604 66.232 53.777 52.339 60.915 50.432 56.049 65.406 67.961 59.828 Rawnet2 [Tak et al., 2021b] Open 17.6 54.004 52.434 50.817 66.960 59.929 59.331 58.781 34.342 49.781 50.641 52.372 44.360 35.447 60.905 51.826 57.016 Table 7: F1-score performance on Speech-DF-Arena ASVspoof ADD System Licence Params Pooled Avg. ITW 19 21-LA 21-DF 24-E FoR CF 22-T1 22-T3 23-R1 23-R2 DFADD LSV SONAR Teffic-Audio Proprietary 590.0 0.969 0.973 0.989 0.989 0.906 0.955 0.966 0.990 0.988 0.884 0.975 0.998 0.987 0.999 1.000 0.998 Modulate-VELMA-2-Synthetic- Voice [Modulate, 2026] Proprietary 316.0 0.966 0.976 0.990 0.986 0.937 0.943 0.991 0.999 0.975 0.915 0.965 0.993 0.988 0.999 0.991 0.993 Resemble-Detect-3B-Omni [Resemble AI, 2026b] Proprietary 3000.0 0.955 0.949 0.989 0.888 0.865 0.905 0.989 0.976 0.957 0.838 0.954 0.976 0.949 0.999 1.000 0.990 Hiya-Authenticity-Verification Multi-v1 [Hiya, 2026] Proprietary 1000.0 0.950 0.954 0.990 0.990 0.950 0.810 0.980 1.000 0.910 0.810 0.960 0.990 0.970 1.000 1.000 1.000 DLMSL-SpeakSure-v0.1 [DLMSL, 2026] Proprietary 658.6 0.874 0.938 0.990 0.998 0.996 0.998 0.733 0.998 0.893 0.763 0.931 0.960 0.942 0.996 0.995 0.999 DF-Raptor [Idiap, 2026] Proprietary 100.0 0.840 0.836 0.972 0.807 0.691 0.605 0.840 0.969 0.785 0.612 0.886 0.950 0.955 0.944 0.787 0.907 Whispeak [Whispeak, 2026] Proprietary 98.9 0.834 0.925 0.990 0.981 0.843 0.625 0.787 0.990 0.994 0.809 0.933 0.982 0.966 0.999 0.998 0.995 DF_Arena_1B_V_1 [Kulkarni et al., 2026] Open 1000.0 0.812 0.886 0.993 0.947 0.804 0.758 0.662 0.971 0.870 0.668 0.936 0.964 0.919 0.999 0.995 0.990 Momenta [Momenta, 2026] Proprietary 350.0 0.808 0.874 0.963 0.944 0.796 0.795 0.709 0.972 0.864 0.627 0.901 0.924 0.922 0.930 0.943 0.953 Syntra Detector [Syntra, 2026] Proprietary 584.0 0.790 0.870 0.968 0.932 0.550 0.729 0.682 0.995 0.981 0.650 0.914 0.974 0.933 0.999 0.960 0.994 DF_Arena_500M_V_1 [Kulkarni et al., 2026] Open 500.0 0.780 0.884 0.986 0.949 0.819 0.620 0.742 0.977 0.900 0.645 0.920 0.947 0.913 0.999 0.996 0.984 MoLEx [A*STAR, 2026] Proprietary 376.4 0.763 0.836 1.000 0.987 0.748 0.743 0.684 0.998 0.559 0.550 0.896 0.920 0.863 0.854 0.989 0.992 Resemble Detect [Resemble AI, 2026a] Proprietary 2112.0 0.759 0.825 0.968 0.939 0.923 0.586 0.677 0.987 0.552 0.593 0.834 0.843 0.789 0.999 0.945 0.974 DF_Arena_100M_V_1 Closed 100.0 0.738 0.837 0.982 0.930 0.708 0.489 0.599 0.926 0.864 0.607 0.851 0.936 0.878 0.999 0.993 0.961 XLSR+SLS [Zhang et al., 2024] Open 340.0 0.704 0.786 0.940 0.989 0.871 0.741 0.638 0.949 0.548 0.528 0.637 0.856 0.847 0.831 0.934 0.778 DF_Arena_100M_V_0 Closed 100.0 0.700 0.800 0.970 0.830 0.640 0.370 0.600 0.940 0.840 0.550 0.820 0.910 0.860 0.990 0.990 0.950 TCM [Truong et al., 2024] Open 319.0 0.694 0.763 0.937 0.991 0.866 0.718 0.637 0.893 0.520 0.490 0.553 0.824 0.834 0.804 0.922 0.761 BiCrossMamba-ST [Kheir et al., 2025] Open 318.2 0.687 0.765 0.936 0.967 0.834 0.699 0.720 0.931 0.501 0.567 0.588 0.774 0.776 0.813 0.929 0.753 Nes2NetX [Liu et al., 2025] Open 317.9 0.684 0.760 0.937 0.994 0.900 0.786 0.590 0.937 0.484 0.522 0.475 0.842 0.867 0.762 0.906 0.715 Wav2Vec2 AASIST [Tak et al., 2022b] Open 317.8 0.651 0.721 0.909 0.989 0.960 0.440 0.677 0.926 0.443 0.560 0.623 0.788 0.840 0.749 0.693 0.574 XLSR Mamba [Tak et al., 2022b] Open 319.0 0.637 0.780 0.946 0.980 0.955 0.744 0.708 0.933 0.528 0.524 0.577 0.836 0.854 0.771 0.926 0.782 Whisper Mesonet [Kawa et al., 2023] Open 7.6 0.596 0.596 0.775 0.769 0.515 0.721 0.583 0.523 0.534 0.480 0.509 0.671 0.648 0.559 0.612 0.448 Wav2Vec2 ECAPA [Kulkarni et al., 2024] Open 324.0 0.527 0.460 0.739 0.328 0.356 0.161 0.640 0.377 0.467 0.398 0.539 0.724 0.718 0.118 0.417 0.388 AASIST [Jung et al., 2022] Open 0.3 0.483 0.514 0.625 0.961 0.607 0.173 0.425 0.784 0.368 0.384 0.397 0.610 0.754 0.358 0.318 0.461 WavLM ECAPA [Kulkarni et al., 2024] Open 102.0 0.475 0.573 0.703 0.964 0.737 0.227 0.537 0.767 0.415 0.420 0.335 0.778 0.759 0.490 0.379 0.615 RawGatST [Tak et al., 2021a] Open 0.4 0.474 0.512 0.532 0.951 0.636 0.155 0.376 0.469 0.378 0.433 0.407 0.701 0.797 0.564 0.272 0.528 RawTFNet [Xiao et al., 2025] Open 0.2 0.395 0.536 0.665 0.915 0.790 0.216 0.335 0.635 0.362 0.424 0.360 0.693 0.771 0.580 0.405 0.487 Hubert ECAPA [Kulkarni et al., 2024] Open 102.0 0.375 0.519 0.666 0.951 0.582 0.258 0.471 0.662 0.415 0.387 0.338 0.592 0.653 0.432 0.377 0.632 Rawnet2 [Tak et al., 2021b] Open 17.6 0.348 0.363 0.565 0.295 0.230 0.075 0.368 0.344 0.376 0.371 0.265 0.533 0.448 0.385 0.235 0.605

Instant research discovery