File size: 12,131 Bytes
667664f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 |
=====================================================================================================
ASVspoof5 Database
Copyright (c) 2024
ASVspoof5 organizing committee
https://www.asvspoof.org/
=====================================================================================================
1. Directory Structure
_______________________
This package includes
|- flac_T_aa.tar ASVspoof 5 training set data
|...
|- flac_T_ae.tar
|
|- flac_D_aa.tar ASVspoof 5 development set data
|...
|- flac_D_ac.tar
|
|- flac_E_aa.tar ASVspoof 5 evaluation set sdata
|...
|- flac_E_aj.tar
|
|- ASVspoof5_protocols.tar.gz ASVspoof 5 protocol and file lists
|
|- LICENSE.txt
|- README.txt
By unpacking flac_T_**.tar, a new directory flac_T will be created, and the training set speech data will be saved in ./flac_T
By unpacking flac_D_**.tar, a new directory flac_D will be created, the development set speech data will be saved in ./flac_D
By unpacking flac_E_**.tar, a new directory flac_E_eval will be created, the evaluation set speech data will be saved in ./flac_E_eval
By unpacking ASVspoof5_protocols.tar.gz, the following files will be extracted
|
|- ASVspoof5.train.tsv list of training set utterances, keys, and other information
|
|- ASVspoof5.dev.track_1.tsv list of development set utterances, keys, and other information.
| It can be used for model validation in track 1 (detection model without ASV)
|
|- ASVspoof5.dev.trial.txt list of ASV trials created from development set data.
| It can be used for model validation in track 2 (Spoofing robust ASV)
|
|- ASVspoof5.dev.enroll.txt list of enrollment utterances of targets speaker
| in the development set
|
|- ASVspoof5.eval.track_1.tsv list of evaluation set utterances, keys, and other information.
| It is used for track 1 evaluation
|
|- ASVspoof5.eval.trial.tsv list of ASV trials created from evaluation set data.
| It is used for track 2 evaluation
|
|- ASVspoof5.eval.enroll.tsv list of enrollment utterances of targets speaker
| in the evaluation set
|
|- ASVspoof5.codec.config.csv configurations of codec
2. Statistics and md5sum
_______________________
md5sum filename number_of_lines
b0cc86b14826a7701b52aad4f53daf9c ASVspoof5_train/flac_T_aa.tar 36500
d05be3f4be7a343fdbdd0ed29fdff2e1 ASVspoof5_train/flac_T_ab.tar 36500
70d1ba4ad75a20aef3dae541fbd321e3 ASVspoof5_train/flac_T_ac.tar 36500
c9bb56af2cc410d98338d74babcb95c9 ASVspoof5_train/flac_T_ad.tar 36500
a3969816982f3e52300d6147ad796df7 ASVspoof5_train/flac_T_ae.tar 36357
df0be44957623991028cce59792beb17 ASVspoof5_dev/flac_D_aa.tar 47400
1e8cd685d89b64502692f1bcf1a13db3 ASVspoof5_dev/flac_D_ab.tar 47400
5e0031f08c30e4bdbf0c59f91b2d662b ASVspoof5_dev/flac_D_ac.tar 47334
a8c800766f3d4ef87971e2b4f29663e2 ASVspoof5_eval/flac_E_aa.tar 68188
c35064188f54f07c87aba58de22534a0 ASVspoof5_eval/flac_E_ab.tar 68188
ed7cbcd1b2847998b72472ddf6b445e3 ASVspoof5_eval/flac_E_ac.tar 68188
626a080e994b05df49c577d7b3dede8d ASVspoof5_eval/flac_E_ad.tar 68188
05f19a5e64fa556714ae1b38eb2ea70b ASVspoof5_eval/flac_E_ae.tar 68188
2ca2f52a3bbf827f7ec155ecb47e85d6 ASVspoof5_eval/flac_E_af.tar 68188
671200b5de2fc1e74a563b07b057af8f ASVspoof5_eval/flac_E_ag.tar 68188
eaf885b1299a61eb07f71e82e96dba37 ASVspoof5_eval/flac_E_ah.tar 68188
b2625683bf440abae2af901a45542ad9 ASVspoof5_eval/flac_E_ai.tar 68188
c50281c8233af6f3b0899604d21ade44 ASVspoof5_eval/flac_E_aj.tar 68180
865d0e894ea9f686f0f37e5ae3ae3616 ASVspoof5_protocols.tar.gz
cbfa08e2fab1423e828f102bc97a22f0 ASVspoof5.train.tsv 182357
e949dee1091a49b04943c548b5cf9860 ASVspoof5.dev.track_1.tsv 140950
11d144e71599f455265bfaa229a3168f ASVspoof5.dev.track_2.enroll.tsv 398 (1184 utt.)
86b20750e7df2c5607401a98225b8606 ASVspoof5.dev.track_2.trial.tsv 282456
50c0591d45a282cb787a5a4614394f67 ASVspoof5.eval.track_1.tsv 680774
fb754cb8905e7d0816cccaca2d8ffd5f ASVspoof5.eval.track_2.enroll.tsv 367 (1098 utt.)
de83d6de35d80b58e932fe4ecfccc6e4 ASVspoof5.eval.track_2.trial.tsv 834536
Note
a. Speech waveform data are in FLAC format (https://en.wikipedia.org/wiki/FLAC).
The sampling rate is 16 kHz.
b. In total,
flac_T_*.tar: 182357 utterances
flac_D_*.tar: 142134 utterances
= 140950 in ASVspoof5.dev.track_1.tsv + 1184 in ASVspoof5.dev.track_2.enroll.tsv
flac_E_*.tar: 681872 utterances
= 680774 in ASVspoof5.eval.track_1.tsv + 1098 in ASVspoof5.eval.track_2.trial.tsv
3. Meta data in txt file
_______________________
ASVspoof5.*.tsv are space-separated
ASVspoof5.train.tsv, ASVspoof5.dev.track_1.tsv and ASVspoof5.eval.track_1.tsv have FIVE columns:
SPEAKER_ID FLAC_FILE_NAME SPEAKER_GENDER CODEC CODEC_Q CODEC_SEED ATTACK_TAG ATTACK_LABEL KEY TMP
SPEAKER_ID: T_****, D_****, or E_**** ID of the speaker in the FLAC file
FLAC_FILE_NAME: T/D/E_********** name of the FLAC file
SPEAKER_GENDER: F or M gender of the speaker
CODEC: C** or - name of the codec or compressor condition
CODEC_Q: N or - codec quality factor configuration number
CODEC_SEED: T/D/E_********** or - if this utterance is coded, name of the original utterance
ATTACK_TAG: AC* or - tag of the attacker adaptation condition
ATTACK_LABEL: A** or bonafide name of the attack
KEY: spoof or bonafide CM key
TMP: - reserved column
ASVspoof5.dev.track_2.trial.tsv and ASVspoof5.eval.track_2.trial.tsv have FIVE columns:
TARGET_SPEAKER_ID FLAC_FILE_NAME TARGET_GENDER ATTACK_LABEL ASV_KEY
TARGET_SPEAKER_ID: D_**** or E_**** ID of the ASV target speaker
FLAC_FILE_NAME: D_********** or E_*********** name of the FLAC file
TARGET_GENDER: F or M gender of the target speaker
ATTACK_LABEL A** or bonafide Label of the attack or bonafide
ASV_KEY: target, nontarget, or spoof ASV label
ASVspoof5.dev.enroll.tsv and ASVspoof5.eval.enroll.tsv have TWO columns:
TARGET_SPEAKER_ID FLAC_FILE_NAMES
TARGET_SPEAKER_ID: D/E_**** ID of the ASV target speaker
FLAC_FILE_NAMES: D/E_******,D/E_*****,D/E_***** name of the FLAC files for enrollment
Note
a. FLAC_FILE_NAMES in ASVspoof5.*.enroll.tsv is list of file names separated by ','.
Some target speakers may have less than 3 enrollment utterances
b. The first column in ASVspoof5.*.track_2.trial.tsv is the ID of the target speaker.
The first column in ASVspoof5.*.track_1.tsv and ASVspoof5.train.tsv is the actual speaker ID in the utterance.
The actual speaker may not be equal to the target speaker when the utterance is used in a
non-target AsV trial
c. ASVspoof5.*.track_2.trial.tsv contains limited information on each probe utterance.
You may retrieve information (e.g., CODEC) from the corresponding ASVspoof5.*.track_1.tsv
d. for codec quality and the corresponding bit rates, please check ASVspoof5.codec.config.csv
4. Copyright and License
________________________
You are free to use this database under Open Data Commons Attribution License (ODC-By).
Regarding Open Data Commons Attribution License (ODC-By), please see
https://opendatacommons.org/licenses/by/1.0/index.html
THIS DATABASE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS DATABASE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
The bona fide data is distributed under CC BY 4.0 Deed | Attribution 4.0 International.
Please see https://creativecommons.org/licenses/by/4.0/
5.Ethics
______________________
ASVspoof 5 is committed to upholding ethical standards and to responsible research practices. Our objective is to enhance the security and reliability of automatic speaker verification (ASV) technology by promoting collaboration and progress in the development of robust spoofing/deepfake detection solutions, to promote participation and to protect the interests of stakeholders.
Data contributors and challenge participants are requested to adhere to local data protection regulations when processing speech data. We ask that you conduct your research and development activities in a responsible manner and be mindful of the potential for the misuse of software solutions and results. You are expected to disclose identified vulnerabilities or weaknesses in ASV technology in a responsible manner. The prompt and appropriate reporting of such findings is the shared responsibility of all in our community, contributes to the improvement of security systems and protects against potential misuse.
The ASVspoof organisers explicitly disassociate themselves from any association with or endorsement of hacking activities, unauthorized access attempts, or the creation of spoofs/deepfakes for malicious purposes or personal gain. We strictly condemn any misuse of the knowledge and tools developed through ASVspoof. Any malicious use of challenge outcomes, results or findings is strictly prohibited.
Please note that the ISCA Code of Ethics, available at https://www.isca-speech.org/iscaweb/index.php/about-isca?id=279, applies to all research publications and reports originating from
the ASVspoof initiative and challenge series.
6. Citation
______________________
Please cite the following papers if you use ASVspoof5 database
PLACEHOLDER_FOR_DATABASE_PAPER
Wang, X., Delgado, H., Tak, H., Jung, J.-w., Shim, H.-j., Todisco, M., Kukanov, I., Liu, X., Sahidullah, M., Kinnunen, T.H., Evans, N., Lee, K.A., Yamagishi, J. (2024) ASVspoof 5: crowdsourced speech data, deepfakes, and adversarial attacks at scale. Proc. The Automatic Speaker Verification Spoofing Countermeasures Workshop (ASVspoof 2024), 1-8, doi: 10.21437/ASVspoof.2024-1
@inproceedings{wang24_asvspoof,
title = {ASVspoof 5: crowdsourced speech data, deepfakes, and adversarial attacks at scale},
author = {Xin Wang and Héctor Delgado and Hemlata Tak and Jee-weon Jung and Hye-jin Shim and Massimiliano Todisco and Ivan Kukanov and Xuechen Liu and Md Sahidullah and Tomi H. Kinnunen and Nicholas Evans and Kong Aik Lee and Junichi Yamagishi},
year = {2024},
booktitle = {The Automatic Speaker Verification Spoofing Countermeasures Workshop (ASVspoof 2024)},
pages = {1--8},
doi = {10.21437/ASVspoof.2024-1},
}
7. Acknowledgements
______________________
This database is based on Multilingual LibriSpeech (MLS) [1].
The spoofed data are created by many contributors, and they will be co-authors of a database-description paper.
[1] V. Pratap, Q. Xu, A. Sriram, G. Synnaeve, and R. Collobert, “MLS: A large-scale multilingual dataset for speech research,” in Proc. Interspeech, 2020. doi: 10.21437/Interspeech.2020-2826.
|