In this paper, we analyze two main factors of Bonafide Resource (BR) or AI-based Generator (AG) which affect the performance and the generality of a Deepfake Speech Detection (DSD) model. To this end, we first propose a deep-learning based model, referred to as the baseline. Then, we conducted experiments on the baseline by which we indicate how Bonafide Resource (BR) and AI-based Generator (AG) factors affect the threshold score used to detect fake or bonafide input audio in the inference process. Given the experimental results, a dataset, which re-uses public Deepfake Speech Detection (DSD) datasets and shows a balance between Bonafide Resource (BR) or AI-based Generator (AG), is proposed. We then train various deep-learning based models on the proposed dataset and conduct cross-dataset evaluation on different benchmark datasets. The cross-dataset evaluation results prove that the balance of Bonafide Resources (BR) and AI-based Generators (AG) is the key factor to train and achieve a general Deepfake Speech Detection (DSD) model.
Figures & tables
Fig. 1: The proposed baseline architecture
Datasets
Years
Speakers
Utt. No.
Fake speech
Real Speech
(Male/Female)
(Bonafide/Fake)
Generators
Resources
ASVspoof 2015 [ 7 ] (audio)
2015
45/61
16,651/246,500
10 (7 VC, 3 TTS)
Speaker Volunteers
FoR [ 8 ] (audio)
2019
140
-/195541
7 TTS
Kaggle [ 9 ]
ASVspoof 2019 (LA task) [ 10 ] (audio)
2019
46/61
12,483/108,978
19 (8VC, 11 TTS)
Speaker Volunteers
DFDC [ 11 ] (video)
2020
3426
128,154/104,500
1 TTS
Speaker Volunteers
ASVspoof 2021 (LA task) [ 12 ] (audio)
2021
21/27
18,452/163,114
13 TTS/VC
Speaker Volunteers
TABLE I: Public and Benchmark Datasets Proposed for Deepfake Speech Detection (DSD) in English
Fig. 2: The statistics of DSD datasets in English
Datasets
Bonafide Utterances
Fake Utterances
AG dataset
53,552
861,304
BR dataset
50,131
273,176
BR-AG dataset
103,683
1,134,408
TABLE II: Proposed datasets to balance AI-Based Generator and Bonafide Resource factors
Fig. 3: Distribution of fake probabilities from In-The-Wild dataset using AG dataset to train the baseline model
Fig. 4: Distribution of Logarit scores from In-The-Wild dataset using AG dataset to train the baseline model
Fig. 5: Distribution of fake probabilities from In-The-Wild dataset using BR dataset to train the baseline model
Fig. 6: Distribution of Logarit scores from In-The-Wild dataset using BR dataset to train the baseline model
Fig. 7: Confusion matrix (%) for In-The-Wild dataset using the AG dataset to train the baseline model (The boundary probability of 0.5)
Fig. 8: Confusion matrix (%) for In-The-Wild dataset using the BR dataset to train the baseline model (The boundary probability of 0.5)
Fig. 9: Our proposed model with three-stage training strategy for DSD task
Model
Train Data
Test Data
Acc.
F1
AuC
WavLM-Large frozen + MLP (baseline)
AG dataset
In-The-Wild
0.71
0.72
0.74
WavLM-Large frozen + MLP (baseline)
BR dataset
In-The-Wild
0.75
0.58
0.69
WavLM-Large frozen + MLP (baseline)
BR-AG dataset
In-The-Wild
0.80
0.72
0.82
WavLM-Large finetune + MLP
BR-AG dataset
In-The-Wild
0.87
0.79
0.83
Whisper-base finetune + MLP
BR-AG dataset
In-The-Wild
0.73
0.60
0.65
Wav2Vec2-XLSR-53 finetune + MLP
BR-AG dataset
In-The-Wild
0.79
0.64
0.73
TABLE III: Evaluation of the BR and AG factors on In-The-Wild dataset (Using Probability Boundary of 0.5)
Test Dataset
Accuracy
F1 Score
AuC
In-The-Wild
0.87
0.79
0.83
ASVspoof5 Eva Subset
0.91
0.91
0.92
FakeAVCeleb
0.99
0.99
0.99
TABLE IV: Cross-Dataset Evaluation of the best model (WavLM-Large finetune + MLP) using the boundary probability of 0.5
Fig. 10: Distribution of fake probabilities from In-The-Wild dataset using BR-AG dataset to train the WavLM-Large-Finetune+MLP model
Fig. 11: Distribution of Logarithmic scores from In-The-Wild dataset using BR-AG dataset to train the WavLM-Large-Finetune+MLP model
Speech deepfake detection is a well-established research field with different models, datasets, and training strategies. However, the lack of standardized implementations and evaluation protocols limits reproducibility, benchmarking, and comparison across studies. In this work, we present DeepFense, a comprehensive, open-source PyTorch toolkit integrating the latest architectures, loss functions, and augmentation pipelines, alongside over 100 recipes. Using DeepFense, we conducted a large-scale evaluation of more than 400 models. Our findings reveal that while carefully curated training data improves cross-domain generalization, the choice of pre-trained front-end feature extractor dominates overall performance variance. Crucially, we show severe biases in high-performing models regarding audio quality, speaker gender, and language. DeepFense is expected to facilitate real-world deployment with the necessary tools to address equitable training data selection and front-end fine-tuning.
Yassine El Kheir, Arnab Das, Yixuan Xiao +6
German Research Center for Artificial Intelligence (DFKI), Germany · University of Stuttgart, Germany · National Institute of Informatics, Japan +1
The modern generative audio models can be used by an adversary in an unlawful manner, specifically, to impersonate other people to gain access to private information. To mitigate this issue, speech deepfake detection (SDD) methods started to evolve. Unfortunately, current SDD methods generally suffer from the lack of generalization to new audio domains and generators. More than that, they lack interpretability, especially human-like reasoning that would naturally explain the attribution of a given audio to the bona fide or spoof class and provide human-perceptible cues. In this paper, we propose HIR-SDD, a novel SDD framework that combines the strengths of Large Audio Language Models (LALMs) with the chain-of-thought reasoning derived from the novel proposed human-annotated dataset. Experimental evaluation demonstrates both the effectiveness of the proposed method and its ability to provide reasonable justifications for predictions.
Claims about the robustness and fairness of deepfake speech detectors are only as credible as the datasets used to train and evaluate those systems. We present a dataset-level audit of the deepfake speech landscape. We compile and analyze 39 deepfake speech datasets, examining key attributes including accessibility, documentation, demographic and language coverage, dataset scale, and the underlying bona fide speech sources. Our audit reveals two important takeaways. Firstly, fairness assessment is largely infeasible because most datasets lack demographic metadata, and only a few contain gender or language labels. This prevents any meaningful subgroup analysis and leaves other demographic attributes unaddressed. Secondly, we identify substantial overlap in underlying bona fide source corpora across datasets, which can undermine cross-dataset evaluation and lead to overstated generalization claims.