Hugging Face's logo

Datasets:
lab260
/
LRLspoof

Tasks:
Audio Classification
Languages:
Abkhaz
Armenian
Assamese
Size:
1M<n<10M
ArXiv:
Tags:
antispoofing
text-to-speech
anti-spoofing
audio-deepfake-detection
speech
benchmark
DOI:
License:
Dataset card Files
xet
 Community
2
korallll commited on
Commit
5caf113
·
verified ·
1 Parent(s): 24eaac6

arena: add eval.yaml + data/labels.parquet (labels-only) + arena README front-matter

Browse files
Files changed (3) hide show
  1. README.md +28 -330
  2. data/labels.parquet +3 -0
  3. eval.yaml +32 -0
README.md CHANGED
@@ -72,342 +72,40 @@ task_categories:
72
  tags:
73
  - antispoofing
74
  - text-to-speech
 
 
 
 
 
75
  pretty_name: LRL spoof
76
  size_categories:
77
  - 1M<n<10M
 
 
 
 
 
 
 
 
78
  ---
79
 
80
- # LRLspoof: Low-Resource Language Spoofing Dataset
81
 
82
- A large-scale speech deepfake / anti-spoofing dataset for **low-resource and multilingual** languages. The dataset contains synthetic (TTS-generated) speech from multiple text-to-speech systems across many languages, intended for training and evaluating audio deepfake detection and anti-spoofing models.
 
83
 
 
 
 
 
84
 
85
- **Languages:** abkhazian, armenian, assamese, australian, avar, azerbaijanian, bashkir, belarusian, bodo, brazilian, bulgarian, catalan, chechen, chuvash, crimean, croatian, czech, danish, dutch, english, erzya, farsi, finnish, french, georgian, gujarati, hindi, hungarian, icelandic, indonesian, italian, japan, kalmyk, karachai_balkar, kazakh, khakassky, korean, kyrgyz, latvian, lezgian, lithuanian, luxembourgish, malasian, malayalam, manipuri, mari, marathi, moksha, nepali, nogai, norwegian, odia, ossetian, polish, rajasthani, russian, slovenian, spanish, swahili, tatar, telugu, turkmen, tuvan, uzbek, welsh, yakut.
 
 
 
86
 
87
- **Models:** aHoTTS, Chatterbox, Crimean Tatar (Qirimtatar) TTS, eSpeak NG, Fast Pitch, F5-TTS, FishAudio S1, Fun-CosyVoice 3.0, IMS-Toucan, Indic-TTS, Kokoro, Matcha-TTS, MeloTTS, MMS TTS, OuteTTS, Parler-TTS, Piper, RHVoice, Silero TTS, SpeechT5, TurkicTTS, XTTS-v2, Zonos.
88
-
89
- ---
90
-
91
- ## Overview
92
-
93
- - **Purpose:** Benchmark and develop anti-spoofing / deepfake detection systems for under-resourced and multilingual settings.
94
- - **Content:** Synthetic speech utterances generated by various TTS models, with transcriptions.
95
- - **Structure:** One directory per **language**, under it one directory per **TTS model** (or model part). Each model folder contains:
96
- - **Audio:** `.wav` files (e.g. `1.wav`, `2.wav`, …).
97
- - **Transcriptions:** a `.txt` file (one sentence per line, aligned with the audio indices where applicable).
98
-
99
- ---
100
-
101
- ## Statistics
102
-
103
- - **Number of languages:** 66
104
- - **Number of TTS systems:** 23
105
- - **Total hours:** 2732.17
106
-
107
-
108
-
109
- Total audio duration (hours) by TTS model:
110
-
111
- ![Duration by TTS model](duration_by_tts_model.png)
112
-
113
- Total audio duration (hours) by language:
114
-
115
- ![Duration by language](duration_by_language.png)
116
-
117
-
118
- **Table 5.** Summary SRR (%) aggregated across languages. *MSRR (All)*: mean SRR across all languages; *PSRR (All)*: pooled SRR across all languages; *MSRR (Low)*: mean SRR across low-resource languages; *PSRR (Low)*: pooled SRR across low-resource languages. **Higher is better** for all metrics.
119
-
120
- | Model | MSRR (All) | PSRR (All) | MSRR (Low) | PSRR (Low) |
121
- |-------|------------|------------|------------|------------|
122
- | [AASIST3](https://huggingface.co/MTUCI/AASIST3) | 90.40 | 91.40 | 93.28 | 94.28 |
123
- | [DF Arena 1B](https://huggingface.co/Speech-Arena-2025/DF_Arena_1B_V_1) | 71.07 | 70.53 | 73.62 | 74.89 |
124
- | [DF Arena 500M](https://huggingface.co/Speech-Arena-2025/DF_Arena_500M_V_1) | 63.47 | 58.50 | 67.31 | 62.47 |
125
- | [Rest2TCNGuard](https://github.com/mtuciru/Res2TCNGuard) | 33.07 | 35.37 | 37.38 | 42.94 |
126
- | [ResCapsGuard](https://github.com/mtuciru/ResCapsGuard) | 41.29 | 40.56 | 40.81 | 40.65 |
127
- | [XSLS](https://github.com/QiShanZhang/SLSforASVspoof-2021-DF) | 36.04 | 34.85 | 39.66 | 39.97 |
128
- | [Wav2Vec+AASIST](https://github.com/TakHemlata/SSL_Anti-spoofing) | 45.62 | 43.48 | 50.09 | 49.34 |
129
- | [TCM-ADD](https://github.com/ductuantruong/tcm_add) | 45.93 | 43.20 | 51.30 | 49.66 |
130
- | [Nes2Net](https://github.com/Liu-Tianchi/Nes2Net) | 34.23 | 30.85 | 38.34 | 35.61 |
131
- | [wav2vec2-xls-r-1b-DeepFake-AI4TRUST](https://huggingface.co/DavidCombei/wav2vec2-xls-r-1b-DeepFake-AI4TRUST) | 26.79 | 26.07 | 30.28 | 30.14 |
132
- | [wav2vec2-xls-r-300m-deepfake-V1](https://huggingface.co/DavidCombei/wav2vec2-xls-r-300m-deepfake-V1) | 80.51 | 78.32 | 83.90 | 81.29 |
133
- | [Spectra-0](https://huggingface.co/MTUCI/spectra_0) | **97.11** | **97.47** | **96.67** | **97.23** |
134
-
135
- ---
136
-
137
- ## Dataset Structure
138
-
139
- ```
140
- lrl_spoof/
141
- ├── <language_1>/ # e.g. english, polish, hindi
142
- │ ├── <model_1>/ # e.g. pipertts, kokoro_part1
143
- │ │ ├── 1.wav
144
- │ │ ├── 2.wav
145
- │ │ └── ...
146
- │ ├── <model_1>.txt # transcriptions for model_1 (when present)
147
- │ ├── <model_2>/
148
- │ └── ...
149
- ├── <language_2>/
150
- └── ...
151
- ```
152
-
153
- - **Path to a given subset:** `lrl_spoof/<language>/<model>/` (audio) and `lrl_spoof/<language>/<model>.txt` (transcriptions, when present).
154
-
155
- ---
156
-
157
- ## Download, Combining Parts, and Unpacking
158
-
159
- The dataset is distributed as a compressed archive split into **15 GB** parts (e.g. `lrl_spoof.tar.gz.part_aa`, `lrl_spoof.tar.gz.part_ab`, …). Download all parts from the dataset page on Hugging Face or ModelScope.
160
-
161
- ### Combining parts
162
-
163
- *[Add the exact filenames and commands you use. Example for Linux/macOS:]*
164
-
165
- ```bash
166
- # Example: if parts are named lrl_spoof.tar.gz.part_aa, .part_ab, ...
167
- cat lrl_spoof.tar.gz.part_* > lrl_spoof.tar.gz
168
- ```
169
-
170
- ### Unpacking
171
-
172
- ```bash
173
- tar -xzf lrl_spoof.tar.gz
174
- ```
175
-
176
-
177
- ---
178
-
179
- ## Contact
180
-
181
- - **Email:** k.n.borodin@mtuci.ru
182
- - **Telegram channel:** https://t.me/korallll_ai
183
-
184
- ---
185
-
186
- ## Citation
187
-
188
- If you use this dataset in your research, please cite the accompanying paper:
189
-
190
- ```bibtex
191
- @misc{borodin2026spoofdetectorstravelevaluation,
192
- title={When Spoof Detectors Travel: Evaluation Across 66 Languages in the Low-Resource Language Spoofing Corpus},
193
- author={Kirill Borodin and Vasiliy Kudryavtsev and Maxim Maslov and Mikhail Gorodnichev and Grach Mkrtchian},
194
- year={2026},
195
- eprint={2603.02364},
196
- archivePrefix={arXiv},
197
- primaryClass={cs.SD},
198
- url={https://arxiv.org/abs/2603.02364},
199
- }
200
- ```
201
-
202
-
203
- ### Checksums
204
-
205
- To ensure that your download is correct and not corrupted, you can verify SHA256 checksums with the `sha256sum` tool (available on Linux and most Unix-like systems).
206
-
207
- ```bash
208
- # Verify the combined archive
209
- sha256sum lrl_spoof.tar.gz
210
-
211
- # Verify all parts (run in the directory containing the parts)
212
- sha256sum lrl_spoof.tar.gz.part_* > lrl_spoof_parts.sha256
213
- cat lrl_spoof_parts.sha256
214
- ```
215
-
216
- After combining parts, verify the full archive with the checksum below. You can also verify each part before combining.
217
-
218
- **Whole archive** (after combining parts into `lrl_spoof.tar.gz`):
219
-
220
- ```
221
- lrl_spoof.tar.gz 624cba39b9c60d35ec90c9ee4450385ad885ca31c093e9a6de592f5f524e203b
222
- ```
223
-
224
- **Per-part checksums** (SHA256 of each part file):
225
-
226
- | Part file | SHA256 |
227
- |-----------|--------|
228
- | `lrl_spoof.tar.gz.part_aa` | ee37cb5a5de1dcc2226d2a1f7d1b986d81529fd0a6ceac354a13f184ec9a8df0 |
229
- | `lrl_spoof.tar.gz.part_ab` | 4704e5020710d4151fe986f510bec861d7ff5cc8e7ee2403c01dca7b0cf9ab74 |
230
- | `lrl_spoof.tar.gz.part_ac` | f4973e4bad14f772e7510bc9f8422f86f4538af017f4fb2b83e62500a52a8054 |
231
- | `lrl_spoof.tar.gz.part_ad` | 18c09449888f20a25346a5362c77c6aece27ffab1d017789c08fa96d8a9e545f |
232
- | `lrl_spoof.tar.gz.part_ae` | 46ed67962bdf454ca506d324a171894b1ab95c48a70e580d112e7aa0a28207c0 |
233
- | `lrl_spoof.tar.gz.part_af` | b0a46aa45c13c54b6f29af2e1482236b785beb9f394fe4c397dd44536890a1be |
234
- | `lrl_spoof.tar.gz.part_ag` | ab26e02c455caa1966edde60057102847b2d4383b63f6c7fbbc2f2d64b262d1b |
235
- | `lrl_spoof.tar.gz.part_ah` | bd92dfdcfae766e3c2a9a3db164a97c573bd72f8caab980861fe7c602d75d8e3 |
236
- | `lrl_spoof.tar.gz.part_ai` | ea7da360febfd76fbd7c024e080576bec0c1a4e95b1a42f193317c9fdbb8f968 |
237
- | `lrl_spoof.tar.gz.part_aj` | 335504f44b2c9e0747621517c752b4329f2ec9e99afe85886fc88ab93b43a147 |
238
- | `lrl_spoof.tar.gz.part_ak` | 8317f9cc2cc7eee4903dcf2238f44a5b13b57ca4bb76a73f475741b195ca05c4 |
239
- | `lrl_spoof.tar.gz.part_al` | 830690aa2d432ee8fd0fd6d8cdf603fc2da4988fe646bc9f5584443b5a9b4c8a |
240
- | `lrl_spoof.tar.gz.part_am` | 578d6206f12e73df83cb0dfd825e6d7b1975f44119d79b00c0039fb419d70767 |
241
- | `lrl_spoof.tar.gz.part_an` | 4d8b8fac6dd72b595e1a8786eb90a15827a8cdab89f3bdb7055d38f9b3c2c9d6 |
242
- | `lrl_spoof.tar.gz.part_ao` | 562e5ff609a1043fa08ee842fc71ab5c73988826acc33e5345631bf208570ef9 |
243
- | `lrl_spoof.tar.gz.part_ap` | d13604b3e9055037e46166e708c19d07d01c8f6392d148d18cc0da85a0e5daf1 |
244
- | `lrl_spoof.tar.gz.part_aq` | d3e61174f56a8b12b9473c2fcc4a0b4e747efc26de685bc1ddcc907ca8632f77 |
245
- | `lrl_spoof.tar.gz.part_ar` | 9240f7715d60edbe369fc4db1ee41422a05163f7ea9c8d00c911aa26d34bf2d3 |
246
- | `lrl_spoof.tar.gz.part_as` | c7e385fc0d8578143b7e15862a2c3fc35f5ab96bdd2b5462448cbf7014eb28de |
247
- | `lrl_spoof.tar.gz.part_at` | e6e6b42db89e06d7e72d2aa8a4046c31019aa38179c56dda38faf911d541fe3e |
248
- | `lrl_spoof.tar.gz.part_au` | d492ddd786d0cb24eabbd16af215e87be2d6480b0d5fabf9f42c59bd5232b1bf |
249
- | `lrl_spoof.tar.gz.part_av` | a8cb2118b3c7b2ed29aef831bd117ae6342b21424a37fb73ad38bba4c246bb75 |
250
- | `lrl_spoof.tar.gz.part_aw` | 4b2f4a9d9ebdbd41be88a296297fd4442c968d717903e5c1b2d867309569b03a |
251
- | `lrl_spoof.tar.gz.part_ax` | 1fb78e4aa022f313b375ef7f6cd1111e0ccdcea303be7baa28230de95937f153 |
252
- | `lrl_spoof.tar.gz.part_ay` | 00fcb87e7ac8c6057409b438c150ec8c0e32ccd1f01ca8f9f8c3ca8620b67f02 |
253
- | `lrl_spoof.tar.gz.part_az` | a93904cf3e9ec2e6e1a0714748286f3215e2d32396ab08831f64f0e04450eae7 |
254
- | `lrl_spoof.tar.gz.part_ba` | 5541d5dd7e2d97b45b25881ace22e6d91359e14a7e7aabdd40f4d8665d9d6dcb |
255
- | `lrl_spoof.tar.gz.part_bb` | 0e853d7dcccee31aa88e9e087bdf909b87afa4cd76fd561505376375cb7ae42e |
256
- | `lrl_spoof.tar.gz.part_bc` | 03ba430fc61596dc4dfe35ddee8dd02d8e9878fa19e34ea8eb99a3d128df8e92 |
257
- | `lrl_spoof.tar.gz.part_bd` | ed72149dcd57d80337148e414a38f7b477314f53ceac31e1c3eaa4d2f4876f35 |
258
- | `lrl_spoof.tar.gz.part_be` | f078fd3816cb3962285a98a2241d1480c9b22e2ff86e15f760cf594abf1d9d2a |
259
- | `lrl_spoof.tar.gz.part_bf` | 517a03f070dfecc2aec3d366c75624b9bc7602f35708e3394e4f65e0c770b0ea |
260
- | `lrl_spoof.tar.gz.part_bg` | 50c8c12fc62c1e702ee07ca01ca793fce604ab06d7ae8ab4f5affbea26d0c32b |
261
- | `lrl_spoof.tar.gz.part_bh` | 754250885d5e901c2898df2149608430d16559f7d02889f66cda514ee7fbe95c |
262
- | `lrl_spoof.tar.gz.part_bi` | 90cb8439c7a966f146f905a94f8969538cd12e95aa304b1d03d31d71a7c12f2a |
263
- | `lrl_spoof.tar.gz.part_bj` | 03a3e8ee7b4b4264ffcf6a28d62863f78b3b360ebf3e80eb20251608f5e10828 |
264
- | `lrl_spoof.tar.gz.part_bk` | 0043a4ec8d26032a9cfc0cdb90786518c4fc523d9d5320caba7ed8b739251219 |
265
- | `lrl_spoof.tar.gz.part_bl` | ad036e4cff44aab51addd99b3f44bbcd3e1b3a43b47da2e78fb2bd7b567c1b85 |
266
- | `lrl_spoof.tar.gz.part_bm` | 6ab7e4e597603d3dda0192f20fead863c3f3461535d2003c4e2ccf5ada3f5690 |
267
- | `lrl_spoof.tar.gz.part_bn` | f9f17e350d23eb74a5cad5d7f4f07e8117d4a09f9fd6a474c9b67f97433c7fd8 |
268
- | `lrl_spoof.tar.gz.part_bo` | 093e585745282be049a5848cf3e599de1f8a80d3ed7068500eb5f0ecf815a7c0 |
269
- | `lrl_spoof.tar.gz.part_bp` | c56dda5dd6d5f250bca14542652c26b57d666fb90d7522f6a4e83d3d055780e4 |
270
- | `lrl_spoof.tar.gz.part_bq` | 13b67216fe92b3b7d1cc1525fde3881a0b1768a0ccacbedbd23aa31b57e0232a |
271
-
272
-
273
-
274
-
275
- ---
276
-
277
- ## Languages and TTS Models
278
-
279
- Each row is one (model, language) subset. Paths are relative to the dataset root. **Each subset may have a different licence** (see the Licence column); use and redistribution of a subset must comply with the licence of the corresponding source TTS model.
280
-
281
- **Licences that restrict distribution:** The following licences limit commercial or unrestricted redistribution. Subsets under these licences may be used only in accordance with their terms (typically non‑commercial use only): **CC BY-NC-4.0** (F5-TTS, MMS TTS), **CC BY-NC-SA 4.0** (FishAudio S1, OuteTTS, Silero TTS, TurkicTTS), and **coqui-public-model-license** (XTTS-v2). Subsets under **GPL-2.0** or **GPL-3.0** (e.g. eSpeak NG, Piper, RHVoice) allow distribution under copyleft terms (derivative works must be GPL-licensed). All other listed licences (e.g. Apache-2.0, MIT, BSD-3-Clause) are permissive and do not restrict distribution for research or open use.
282
-
283
- | Model | Language | Path in dataset | Link to model | Licence |
284
- |-------|----------|-----------------|----------------|------|
285
- | aHoTTS | catalan | `lrl_spoof/catalan/ahotts/` | https://github.com/hitz-zentroa/aHoTTS| Apache-2.0 |
286
- | Chatterbox | finnish | `lrl_spoof/finnish/chatterbox/` |https://github.com/resemble-ai/chatterbox | MIT |
287
- | Chatterbox | malasian | `lrl_spoof/malasian/chatterbox/` |https://github.com/resemble-ai/chatterbox | MIT |
288
- | Chatterbox | malayalam | `lrl_spoof/malayalam/chatterbox/` |https://github.com/resemble-ai/chatterbox | MIT |
289
- | Chatterbox | norwegian | `lrl_spoof/norwegian/chatterbox_part1/` |https://github.com/resemble-ai/chatterbox | MIT |
290
- | Chatterbox | norwegian | `lrl_spoof/norwegian/chatterbox_part2/` |https://github.com/resemble-ai/chatterbox | MIT |
291
- | Chatterbox | polish | `lrl_spoof/polish/chatterbox_part1/` |https://github.com/resemble-ai/chatterbox | MIT |
292
- | Chatterbox | polish | `lrl_spoof/polish/chatterbox_part2/` |https://github.com/resemble-ai/chatterbox | MIT |
293
- | Chatterbox | swahili | `lrl_spoof/swahili/chatterbox/` |https://github.com/resemble-ai/chatterbox | MIT |
294
- | Fun-CosyVoice 3.0 | japan | `lrl_spoof/japan/cosyvoice3/` |https://github.com/FunAudioLLM/CosyVoice |Apache-2.0 |
295
- | Fun-CosyVoice 3.0 | korean | `lrl_spoof/korean/cosyvoice3/` | https://github.com/FunAudioLLM/CosyVoice|Apache-2.0 |
296
- | eSpeak NG | armenian | `lrl_spoof/armenian/espeak/` |https://github.com/espeak-ng/espeak-ng | GPL-3.0 |
297
- | eSpeak NG | azerbaijanian | `lrl_spoof/azerbaijanian/espeak/` |https://github.com/espeak-ng/espeak-ng |GPL-3.0 |
298
- | eSpeak NG | bashkir | `lrl_spoof/bashkir/espeak/` |https://github.com/espeak-ng/espeak-ng |GPL-3.0 |
299
- | eSpeak NG | bulgarian | `lrl_spoof/bulgarian/espeak_part1/` |https://github.com/espeak-ng/espeak-ng |GPL-3.0 |
300
- | eSpeak NG | bulgarian | `lrl_spoof/bulgarian/espeak_part2/` |https://github.com/espeak-ng/espeak-ng |GPL-3.0 |
301
- | eSpeak NG | chuvash | `lrl_spoof/chuvash/espeak/` |https://github.com/espeak-ng/espeak-ng |GPL-3.0 |
302
- | eSpeak NG | georgian | `lrl_spoof/georgian/espeak/` |https://github.com/espeak-ng/espeak-ng |GPL-3.0 |
303
- | eSpeak NG | kazakh | `lrl_spoof/kazakh/espeak/` |https://github.com/espeak-ng/espeak-ng |GPL-3.0 |
304
- | eSpeak NG | kyrgyz | `lrl_spoof/kyrgyz/espeak/` |https://github.com/espeak-ng/espeak-ng |GPL-3.0 |
305
- | eSpeak NG | turkmen | `lrl_spoof/turkmen/espeak/` |https://github.com/espeak-ng/espeak-ng |GPL-3.0 |
306
- | IMS-Toucan | abkhazian | `lrl_spoof/abkhazian/ims_toucan/` | https://github.com/DigitalPhonetics/IMS-Toucan| Apache-2.0 |
307
- | IMS-Toucan | armenian | `lrl_spoof/armenian/ims_toucan/` |https://github.com/DigitalPhonetics/IMS-Toucan |Apache-2.0 |
308
- | IMS-Toucan | azerbaijanian | `lrl_spoof/azerbaijanian/ims_toucan/` | https://github.com/DigitalPhonetics/IMS-Toucan|Apache-2.0 |
309
- | IMS-Toucan | bashkir | `lrl_spoof/bashkir/ims_toucan/` |https://github.com/DigitalPhonetics/IMS-Toucan |Apache-2.0 |
310
- | IMS-Toucan | chuvash | `lrl_spoof/chuvash/ims_toucan/` |https://github.com/DigitalPhonetics/IMS-Toucan |Apache-2.0 |
311
- | IMS-Toucan | czech | `lrl_spoof/czech/ims_toucan/` |https://github.com/DigitalPhonetics/IMS-Toucan |Apache-2.0 |
312
- | IMS-Toucan | karachai_balkar | `lrl_spoof/karachai_balkar/ims_toucan/` |https://github.com/DigitalPhonetics/IMS-Toucan |Apache-2.0 |
313
- | IMS-Toucan | lezgian | `lrl_spoof/lezgian/ims_toucan/` |https://github.com/DigitalPhonetics/IMS-Toucan |Apache-2.0 |
314
- | IMS-Toucan | tuvan | `lrl_spoof/tuvan/ims_toucan/` |https://github.com/DigitalPhonetics/IMS-Toucan |Apache-2.0 |
315
- | Indic-TTS | assamese | `lrl_spoof/assamese/indictts/` |https://github.com/AI4Bharat/Indic-TTS |MIT |
316
- | Indic-TTS | bodo | `lrl_spoof/bodo/indictts/` |https://github.com/AI4Bharat/Indic-TTS |MIT |
317
- | Indic-TTS | gujarati | `lrl_spoof/gujarati/indictts/` |https://github.com/AI4Bharat/Indic-TTS |MIT |
318
- | Indic-TTS | hindi | `lrl_spoof/hindi/indictts/` |https://github.com/AI4Bharat/Indic-TTS |MIT |
319
- | Indic-TTS | malayalam | `lrl_spoof/malayalam/indictts/` |https://github.com/AI4Bharat/Indic-TTS |MIT |
320
- | Indic-TTS | manipuri | `lrl_spoof/manipuri/indictts/` |https://github.com/AI4Bharat/Indic-TTS |MIT |
321
- | Indic-TTS | marathi | `lrl_spoof/marathi/indictts_part1/` |https://github.com/AI4Bharat/Indic-TTS |MIT |
322
- | Indic-TTS | marathi | `lrl_spoof/marathi/indictts_part2/` |https://github.com/AI4Bharat/Indic-TTS |MIT |
323
- | Indic-TTS | odia | `lrl_spoof/odia/indictts/` |https://github.com/AI4Bharat/Indic-TTS |MIT |
324
- | Indic-TTS | rajasthani | `lrl_spoof/rajasthani/indictts/` |https://github.com/AI4Bharat/Indic-TTS |MIT |
325
- | Indic-TTS | telugu | `lrl_spoof/telugu/indictts/` |https://github.com/AI4Bharat/Indic-TTS |MIT |
326
- | Fast Pitch | english | `lrl_spoof/english/fastpitch/` |https://github.com/dan-wells/fastpitch | BSD-3-Clause |
327
- | F5-TTS | chuvash | `lrl_spoof/chuvash/f5/` | https://huggingface.co/Misha24-10/F5-TTS_CHUVASH | CC BY-NC-4.0 |
328
- | FishAudio S1 | polish | `lrl_spoof/polish/fishspeech/` |https://huggingface.co/fishaudio/s1-mini | CC BY-NC-SA-4.0 |
329
- | Kokoro | brazilian | `lrl_spoof/brazilian/kokoro/` |https://github.com/hexgrad/kokoro |Apache-2.0 |
330
- | Kokoro | english | `lrl_spoof/english/kokoro_part1/` |https://github.com/hexgrad/kokoro |Apache-2.0 |
331
- | Kokoro | english | `lrl_spoof/english/kokoro_part2/` |https://github.com/hexgrad/kokoro |Apache-2.0 |
332
- | Kokoro | french | `lrl_spoof/french/kokoro_part1/` |https://github.com/hexgrad/kokoro |Apache-2.0 |
333
- | Kokoro | french | `lrl_spoof/french/kokoro_part2/` |https://github.com/hexgrad/kokoro |Apache-2.0 |
334
- | Kokoro | french | `lrl_spoof/french/kokoro_part3/` |https://github.com/hexgrad/kokoro |Apache-2.0 |
335
- | Kokoro | hindi | `lrl_spoof/hindi/kokoro/` |https://github.com/hexgrad/kokoro |Apache-2.0 |
336
- | Kokoro | italian | `lrl_spoof/italian/kokoro/` |https://github.com/hexgrad/kokoro |Apache-2.0 |
337
- | Kokoro | japan | `lrl_spoof/japan/kokoro/` |https://github.com/hexgrad/kokoro |Apache-2.0 |
338
- | Matcha-TTS | kyrgyz | `lrl_spoof/kyrgyz/matcha/` |https://huggingface.co/kyrgyz-ai/akylai-tts-mini | MIT |
339
- | MeloTTS | australian | `lrl_spoof/australian/melotts/` | https://github.com/myshell-ai/MeloTTS| MIT |
340
- | MeloTTS | english | `lrl_spoof/english/melotts/` |https://github.com/myshell-ai/MeloTTS | MIT |
341
- | MeloTTS | japan | `lrl_spoof/japan/melotts/` |https://github.com/myshell-ai/MeloTTS | MIT |
342
- | MeloTTS | korean | `lrl_spoof/korean/melotts/` | https://github.com/myshell-ai/MeloTTS| MIT |
343
- | MeloTTS | spanish | `lrl_spoof/spanish/melotts/` |https://github.com/myshell-ai/MeloTTS | MIT |
344
- | MMS TTS | armenian | `lrl_spoof/armenian/mms_tts/` | https://huggingface.co/facebook/mms-tts-hyw| CC BY-NC-4.0 |
345
- | MMS TTS | azerbaijanian | `lrl_spoof/azerbaijanian/mms_tts/` |https://huggingface.co/facebook/mms-tts-azb | CC BY-NC-4.0 |
346
- | MMS TTS | bashkir | `lrl_spoof/bashkir/mms_tts/` |https://huggingface.co/facebook/mms-tts-bak |CC BY-NC-4.0 |
347
- | MMS TTS | chuvash | `lrl_spoof/chuvash/mms_tts/` |https://huggingface.co/facebook/mms-tts-chv |CC BY-NC-4.0 |
348
- | MMS TTS | dutch | `lrl_spoof/dutch/mms_tts/` |https://huggingface.co/facebook/mms-tts-nld |CC BY-NC-4.0 |
349
- | MMS TTS | latvian | `lrl_spoof/latvian/mms_tts/` |https://huggingface.co/facebook/mms-tts-lav | CC BY-NC-4.0 |
350
- | MMS TTS | yakut | `lrl_spoof/yakut/mms_tts/` |https://huggingface.co/facebook/mms-tts-sah | CC BY-NC-4.0 |
351
- | OuteTTS | belarusian | `lrl_spoof/belarusian/outetts/` |https://huggingface.co/unsloth/Llama-OuteTTS-1.0-1B | CC BY-NC-SA-4.0|
352
- | OuteTTS | georgian | `lrl_spoof/georgian/outetts/` |https://huggingface.co/unsloth/Llama-OuteTTS-1.0-1B | CC BY-NC-SA-4.0|
353
- | OuteTTS | hungarian | `lrl_spoof/hungarian/outetts/` | https://huggingface.co/unsloth/Llama-OuteTTS-1.0-1B| CC BY-NC-SA-4.0|
354
- | OuteTTS | lithuanian | `lrl_spoof/lithuanian/outetts/` | https://huggingface.co/unsloth/Llama-OuteTTS-1.0-1B| CC BY-NC-SA-4.0|
355
- | OuteTTS | polish | `lrl_spoof/polish/outetts/` |https://huggingface.co/unsloth/Llama-OuteTTS-1.0-1B | CC BY-NC-SA-4.0|
356
- | OuteTTS | russian | `lrl_spoof/russian/outetts/` |https://huggingface.co/unsloth/Llama-OuteTTS-1.0-1B | CC BY-NC-SA-4.0|
357
- | Parler-TTS | english | `lrl_spoof/english/parlertts_part1/` | https://huggingface.co/parler-tts/parler-tts-mini-multilingual-v1.1|Apache-2.0 |
358
- | Parler-TTS | english | `lrl_spoof/english/parlertts_part2/` |https://huggingface.co/parler-tts/parler-tts-mini-multilingual-v1.1 |Apache-2.0 |
359
- | Parler-TTS | english | `lrl_spoof/english/parlertts_part3/` |https://huggingface.co/parler-tts/parler-tts-mini-multilingual-v1.1 |Apache-2.0 |
360
- | Parler-TTS | french | `lrl_spoof/french/parlertts_part1/` | https://huggingface.co/parler-tts/parler-tts-mini-multilingual-v1.1|Apache-2.0 |
361
- | Parler-TTS | french | `lrl_spoof/french/parlertts_part2/` | https://huggingface.co/parler-tts/parler-tts-mini-multilingual-v1.1 |Apache-2.0 |
362
- | Parler-TTS | polish | `lrl_spoof/polish/parlertts/` |https://huggingface.co/parler-tts/parler-tts-mini-multilingual-v1.1 |Apache-2.0 |
363
- | Piper | catalan | `lrl_spoof/catalan/pipertts/` |https://github.com/OHF-Voice/piper1-gpl | GPL-3.0 |
364
- | Piper | czech | `lrl_spoof/czech/pipertts/` |https://github.com/OHF-Voice/piper1-gpl | GPL-3.0 |
365
- | Piper | danish | `lrl_spoof/danish/pipertts/` |https://github.com/OHF-Voice/piper1-gpl | GPL-3.0 |
366
- | Piper | farsi | `lrl_spoof/farsi/pipertts/` |https://github.com/OHF-Voice/piper1-gpl | GPL-3.0 |
367
- | Piper | georgian | `lrl_spoof/georgian/pipertts/` |https://github.com/OHF-Voice/piper1-gpl | GPL-3.0 |
368
- | Piper | icelandic | `lrl_spoof/icelandic/pipertts_part1/` |https://github.com/OHF-Voice/piper1-gpl |GPL-3.0 |
369
- | Piper | icelandic | `lrl_spoof/icelandic/pipertts_part2/` |https://github.com/OHF-Voice/piper1-gpl |GPL-3.0 |
370
- | Piper | icelandic | `lrl_spoof/icelandic/pipertts_part3/` |https://github.com/OHF-Voice/piper1-gpl |GPL-3.0 |
371
- | Piper | indonesian | `lrl_spoof/indonesian/pipertts/` |https://github.com/OHF-Voice/piper1-gpl |GPL-3.0 |
372
- | Piper | kazakh | `lrl_spoof/kazakh/pipertts_part1/` |https://github.com/OHF-Voice/piper1-gpl |GPL-3.0 |
373
- | Piper | kazakh | `lrl_spoof/kazakh/pipertts_part2/` | https://github.com/OHF-Voice/piper1-gpl|GPL-3.0 |
374
- | Piper | luxembourgish | `lrl_spoof/luxembourgish/pipertts/` | https://github.com/OHF-Voice/piper1-gpl|GPL-3.0 |
375
- | Piper | nepali | `lrl_spoof/nepali/pipertts/` |https://github.com/OHF-Voice/piper1-gpl |GPL-3.0 |
376
- | Piper | polish | `lrl_spoof/polish/pipertts/` |https://github.com/OHF-Voice/piper1-gpl |GPL-3.0 |
377
- | Piper | slovenian | `lrl_spoof/slovenian/piper/` |https://github.com/OHF-Voice/piper1-gpl |GPL-3.0 |
378
- | Piper | welsh | `lrl_spoof/welsh/pipertts/` |https://github.com/OHF-Voice/piper1-gpl |GPL-3.0 |
379
- | Crimean Tatar (Qirimtatar) TTS | crimean | `lrl_spoof/crimean/qirimtatar_tts/` | https://github.com/robinhad/qirimtatar-tts| MIT |
380
- | RHVoice | kyrgyz | `lrl_spoof/kyrgyz/rhvoice/` |https://github.com/RHVoice/RHVoice | GPL-2.0 |
381
- | Silero TTS | avar | `lrl_spoof/avar/silero/` |https://github.com/snakers4/silero-models | CC BY-NC-SA 4.0 |
382
- | Silero TTS | bulgarian | `lrl_spoof/bulgarian/silero_part1/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
383
- | Silero TTS | bulgarian | `lrl_spoof/bulgarian/silero_part2/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
384
- | Silero TTS | bulgarian | `lrl_spoof/bulgarian/silero_part3/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
385
- | Silero TTS | chechen | `lrl_spoof/chechen/silero_part1/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
386
- | Silero TTS | chechen | `lrl_spoof/chechen/silero_part2/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
387
- | Silero TTS | chuvash | `lrl_spoof/chuvash/silero_part1/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
388
- | Silero TTS | chuvash | `lrl_spoof/chuvash/silero_part2/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
389
- | Silero TTS | chuvash | `lrl_spoof/chuvash/silero_part3/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
390
- | Silero TTS | erzya | `lrl_spoof/erzya/silero/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
391
- | Silero TTS | kalmyk | `lrl_spoof/kalmyk/silero/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
392
- | Silero TTS | khakassky | `lrl_spoof/khakassky/silero/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
393
- | Silero TTS | mari | `lrl_spoof/mari/silero/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
394
- | Silero TTS | moksha | `lrl_spoof/moksha/silero_part1/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
395
- | Silero TTS | moksha | `lrl_spoof/moksha/silero_part2/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
396
- | Silero TTS | moksha | `lrl_spoof/moksha/silero_part3/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
397
- | Silero TTS | nogai | `lrl_spoof/nogai/silero/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
398
- | Silero TTS | ossetian | `lrl_spoof/ossetian/silero/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
399
- | Silero TTS | yakut | `lrl_spoof/yakut/silero/` |https://github.com/snakers4/silero-models |CC BY-NC-SA 4.0 |
400
- | SpeechT5 | croatian | `lrl_spoof/croatian/speecht5_part1/` |https://huggingface.co/nikolab/speecht5_tts_hr | MIT |
401
- | SpeechT5 | croatian | `lrl_spoof/croatian/speecht5_part2/` |https://huggingface.co/nikolab/speecht5_tts_hr | MIT |
402
- | SpeechT5 | uzbek | `lrl_spoof/uzbek/speecht5/` |https://huggingface.co/Inomjonov/speecht5-finetuned-uzbek-1102 | MIT |
403
- | TurkicTTS | azerbaijanian | `lrl_spoof/azerbaijanian/turkic_tts/` | https://github.com/IS2AI/TurkicTTS|CC BY-NC-SA 4.0 |
404
- | TurkicTTS | bashkir | `lrl_spoof/bashkir/turkic_tts/` |https://github.com/IS2AI/TurkicTTS |CC BY-NC-SA 4.0 |
405
- | TurkicTTS | kazakh | `lrl_spoof/kazakh/turkic_tts/` |https://github.com/IS2AI/TurkicTTS |CC BY-NC-SA 4.0 |
406
- | TurkicTTS | tatar | `lrl_spoof/tatar/turkic_tts/` |https://github.com/IS2AI/TurkicTTS |CC BY-NC-SA 4.0 |
407
- | TurkicTTS | uzbek | `lrl_spoof/uzbek/turkic_tts/` |https://github.com/IS2AI/TurkicTTS |CC BY-NC-SA 4.0 |
408
- | TurkicTTS | yakut | `lrl_spoof/yakut/turkic_tts/` |https://github.com/IS2AI/TurkicTTS |CC BY-NC-SA 4.0 |
409
- | XTTS-v2 | czech | `lrl_spoof/czech/xtts/` |https://huggingface.co/coqui/XTTS-v2 | coqui-public-model-license|
410
- | XTTS-v2 | hungarian | `lrl_spoof/hungarian/xtts2/` | https://huggingface.co/coqui/XTTS-v2| coqui-public-model-license |
411
- | XTTS-v2 | italian | `lrl_spoof/italian/xtts2/` |https://huggingface.co/coqui/XTTS-v2 | coqui-public-model-license |
412
- | XTTS-v2 | russian | `lrl_spoof/russian/xtts/` | https://huggingface.co/coqui/XTTS-v2| coqui-public-model-license |
413
- | Zonos | japan | `lrl_spoof/japan/zonos/` |https://github.com/Zyphra/Zonos | Apache-2.0 |
 
72
  tags:
73
  - antispoofing
74
  - text-to-speech
75
+ - anti-spoofing
76
+ - audio-deepfake-detection
77
+ - speech
78
+ - benchmark
79
+ - arena-ready
80
  pretty_name: LRL spoof
81
  size_categories:
82
  - 1M<n<10M
83
+ configs:
84
+ - config_name: default
85
+ data_files:
86
+ - split: test
87
+ path: lrl_spoof.tar.gz.part_*
88
+ viewer: false
89
+ arxiv:
90
+ - '2603.02364'
91
  ---
92
 
93
+ # LRLspoof
94
 
95
+ Multilingual low-resource-language anti-spoofing benchmark: 1,304,455 TTS (spoof)
96
+ utterances across 66 languages and many TTS systems. **Spoof-only** (no bonafide).
97
 
98
+ The audio ships as the multi-part tarball `lrl_spoof.tar.gz.part_*`; download +
99
+ extract it into a `<language>/<tts_model>/<file>.wav` tree. `load_dataset` is not
100
+ the access path here (the dataset is the tarball). `data/labels.parquet`
101
+ (utterance_id + label, all spoof) exists only for arena re-verification.
102
 
103
+ ## Arena scoring
104
+ Scored on **1-SRR** (`srr_complement`, lower is better) at a fixed threshold t*
105
+ **calibrated on DeepVoice**. A submission must include a `calibration` block with
106
+ the DeepVoice operating-point threshold.
107
 
108
+ ## utterance_id convention
109
+ `utterance_id` = the audio file path relative to the dataset root, e.g.
110
+ `english/fastpitch/line_59.wav`. A submitter's `scores.txt` keys by this id;
111
+ emit `<utterance_id> <score>` with higher score = more bonafide.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
data/labels.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:878435e1a689e5884fdb5e613c56a11774a3ce6903e40ea19575d2e752def11b
3
+ size 6345951
eval.yaml ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: LRLspoof
2
+ description: >
3
+ LRLspoof - a large multilingual low-resource-language anti-spoofing benchmark
4
+ of 1,304,455 text-to-speech (spoof) utterances spanning 66 languages and many
5
+ TTS systems. It is SPOOF-ONLY (no bonafide), so it is scored with the 1-SRR
6
+ metric (srr_complement): the fraction of spoof NOT rejected at a fixed
7
+ operating threshold t* transferred from DeepVoice (lower is better). Every
8
+ model must carry a DeepVoice submission to derive t* (placed in the
9
+ submission's `calibration` block). The 500+ GB audio is NOT stored in this
10
+ arena entry - it lives as a multi-part tarball in the source repo; only
11
+ data/labels.parquet (utterance_id -> label, every label = 1 = spoof) is
12
+ shipped here. utterance_id = the audio file's path relative to the dataset
13
+ root, e.g. "english/fastpitch/line_59.wav".
14
+ evaluation_framework: inspect-ai
15
+
16
+ tasks:
17
+ - id: antispoofing_eval
18
+ config: default
19
+ split: test
20
+
21
+ field_spec:
22
+ input: audio
23
+ target: label
24
+
25
+ solvers:
26
+ - name: speech_spoof_bench_solver
27
+
28
+ scorers:
29
+ - name: speech_spoof_scorer
30
+
31
+ metrics:
32
+ - srr_complement