Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 72 of the 72 references it cites were found in a registry or an index; 2 of the 13 sources its claims cite were checked for support.

MLAAD: The Multi-Language Audio Anti-Spoofing Dataset

Open access

Abstract

Text-to-Speech (TTS) technology brings significant advantages, such as giving a voice to those with speech impairments, but also enables audio deepfakes and spoofs. The former mislead individuals and may propagate misinformation, while the latter undermine voice biometric security systems. AI-based detection can help to address these challenges by automatically differentiating between genuine and fabricated voice recordings. However, these models are only as good as their training data, which currently is severely limited due to an overwhelming concentration on English and Chinese audio in anti-spoofing databases, thus restricting its worldwide effectiveness.In response, this paper presents the Multi-Language Audio Anti-Spoof Dataset (MLAAD), created using 52 TTS models, comprising 22 different architectures, to generate 160.2 hours of synthetic voice in 23 different languages. We train and evaluate three state-of-the-art deepfake detection models with MLAAD, and observe that MLAAD demonstrates superior performance over comparable datasets like InTheWild or FakeOrReal when used as a training resource. Furthermore, in comparison with the renowned ASVspoof 2019 dataset, MLAAD proves to be a complementary resource. In tests across eight datasets, MLAAD and ASVspoof 2019 alternately outperformed each other, both excelling on four datasets.By publishing1MLAAD and making trained models accessible via an interactive webserver2, we aim to democratize antispoofing technology, making it accessible beyond the realm of specialists, thus contributing to global efforts against audio spoofing and deepfakes.

Overview

CoChat Summary

TL;DR

MLAAD is a multilingual synthetic-audio anti-spoofing dataset, version 11, containing 1153.6 hours of synthetic voice from 205 TTS models across 54 languages, built as an extension of M-AILABS. Training three state-of-the-art detectors on MLAAD shows cross-dataset generalization that is complementary to ASVspoof 2019, with MLAAD and ASVspoof 2019 each achieving the highest accuracy on four of eight test datasets.

Key findings

  • MLAAD v11 comprises 1153.6 hours of synthesized speech, 205 state-of-the-art TTS models, 127 different architectures, and 54 languages; only synthesized samples are distributed, with originals referred to the M-AILABS dataset.
  • In cross-dataset evaluation, MLAAD and ASVspoof 2019 alternately outperformed each other, each excelling on four of eight datasets; FakeOrReal and InTheWild did not top any case.
  • Models trained on MLAAD were the only ones that did not exhibit performance significantly below 50%, hinting at possibly less detrimental shortcut features in MLAAD than in other datasets.
  • Example results: SSL W2V2 trained on MLAAD achieved 78.0±15.3 on ASVspoof19, 73.7±10.4 on ASVspoof21-LA, 66.0±34.2 on Voc.v, and 69.8±38.4 on WaveFake; WhisperDF trained on MLAAD achieved 86.7±29.4 on Voc.v and 97.2±3.3 on WaveFake.

Methods

  • Dataset construction: MLAAD extends M-AILABS, which has eight original languages (English, French, German, Italian, Polish, Russian, Spanish, Ukrainian). For each language/TTS-model combination, a random subset of 1000 instances is selected; if the target language is absent from M-AILABS, English texts are translated via Neural Machine Translation. Synthetic versions are generated with TTS models from Coqui.ai and Hugging Face, plus Griffin Lim resynthesis for original languages. Audio is stored at 22,050 Hz, 16-bit depth, WAV format, with meta.csv files.
  • Evaluation: three state-of-the-art voice anti-spoofing models—RawGat-ST, SSL-W2V2, and WhisperDF—are trained separately on MLAAD v1, ASVspoof19, FakeOrReal, and In-The-Wild, then evaluated on eight datasets including ASVspoof21-DF, ASVspoof21-LA, Voc.v, and WaveFake. Each configuration is run five times, reporting mean accuracy and standard deviation. Training uses noise/music augmentation and codec encoding, with balanced undersampling per epoch, 100 epochs, and early stopping based on training accuracy.
  • Accuracy is used instead of equal error rate (EER) because some test datasets, such as Voc.v, consist only of spoof samples, and accuracy is argued to better suit real-world actionable output.

Limitations

  • Stated: evaluation of multilingual anti-spoofing capabilities remains future work because there are no other comparable multilingual datasets against which to evaluate MLAAD; strong results on WaveFake, which includes English and Japanese, suggest models profit from MLAAD’s language diversity.
  • Stated: the reported results use MLAAD v1; the evaluation was not re-run for subsequent releases, though later versions are expected to perform substantially better given much larger volume and diversity.
  • Stated/evident: models sometimes learn shortcut features that transfer well to some datasets but fail completely on others, with some accuracy scores below 50%; MLAAD-trained models are less affected but the issue is not eliminated.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/ijcnn60899.2024.10650962

Cites

72 of 72 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

0 Valid · 5 Not valid · 7 Not checked · 30 Own results and interpretations

21 more claims

Create a free account to see all 42 claims and to verify the 2 references still open.

See all claims — free account
  • Probably not a substantial claim. 11
  • It may state two separate claims as one. 7

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 66 references

Research works printed but not linked: 1

  1. 10.21437/Odyssey.2022-16 · DOI · Paper page

Cited by

See all 73

Metadata

Authors
Müller, N. M., Kawa, P., Choong, W. H., Casanova, E., Gölge, E., Müller, T., Syga, P., Sperl, P., & Böttinger, K.
Affiliations
  • Fraunhofer Institute for Applied and Integrated Security (DE)
  • Wrocław University of Science and Technology (PL)
  • Coqui.ai
  • Thorsten-Voice
Venue
2024 International Joint Conference on Neural Networks (IJCNN)
Publisher
IEEE
Published
2024-06-30
Pages
1-7
ISSN
2161-4407 (ISSN-L)
OpenAlex
W4402351686
Cited by
73
References
66
Citations by year
2026: 21 · 2025: 41 · 2024: 11
FWCI
21.58
Type
conference paper
Language
en
Open access
Locations
  • International Joint Conference on Neural Networks (IJCNN) (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • Fraunhofer-Publica (Fraunhofer-Gesellschaft) (submittedVersion)
  • arXiv (Cornell University)
Funders
Ministry of Economic Affairs
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Speech Recognition and Synthesis
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Media Forensic Detection
  • Physical Sciences › Computer Science › Signal Processing › Speech and Audio Processing
Keywords
audio deepfake detection, text-to-speech, multilingual speech, misinformation, ASVspoof 2019

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: MLAAD: The Multi-Language Audio Anti-Spoofing Dataset. https://app.cochat.ai/papers/10.1109/ijcnn60899.2024.10650962
The paper
Müller, N. M., Kawa, P., Choong, W. H., Casanova, E., Gölge, E., Müller, T., Syga, P., Sperl, P., & Böttinger, K. (2024). MLAAD: The Multi-Language Audio Anti-Spoofing Dataset. 2024 International Joint Conference on Neural Networks (IJCNN). https://doi.org/10.1109/ijcnn60899.2024.10650962

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.