vdivyasharma commited on
Commit
cfb0295
Β·
verified Β·
1 Parent(s): a086877

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +7 -9
README.md CHANGED
@@ -62,7 +62,6 @@ dataset_info:
62
 
63
  # IndicSynth
64
  *A Large-Scale Multilingual Synthetic Speech Dataset for Low-Resource Indian Languages*
65
-
66
  **πŸ† Outstanding Paper Award, ACL 2025**
67
 
68
  ---
@@ -71,9 +70,9 @@ dataset_info:
71
 
72
  **IndicSynth** is a novel multilingual synthetic speech dataset designed to advance multilingual **audio deepfake detection (ADD)** and **anti-spoofing** research. It covers **12 low-resource Indian languages** and provides both **mimicry** and **diversity** subsets.
73
 
74
- > 4,000+ hours of synthetic audio
75
- > 12 languages: Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Sanskrit, Tamil, Telugu, Urdu
76
- > Useful for ADD, speaker verification (SV), and bias studies
77
  ---
78
  ## πŸ“‚ Dataset Structure
79
 
@@ -108,8 +107,8 @@ Synthetic data was generated using:
108
  | vits | TTS | Yes | No |
109
  | freevc24 | VC | No | No |
110
 
111
- > **Mimicry subset**: For anti-spoofing research
112
- > **Diversity subset**: Contains diverse set of realistic synthetic voices for multilingual audio deepfake detection research
113
  For more details, please see the Table 1 and Section 3 of our paper: https://aclanthology.org/2025.acl-long.1070.pdf
114
  ---
115
 
@@ -163,7 +162,6 @@ doi = {10.1609/aaai.v37i11.26521}
163
  For questions or feedback, please feel free to reach out at divyas@iiitd.ac.in.
164
 
165
  ## πŸ™ Acknowledgments
166
- > 🌍 ACL Diversity & Inclusion Subsidy for enabling in-person presentation at ACL 2025
167
-
168
- > 🀝 HuggingFace for dataset hosting support
169
 
 
62
 
63
  # IndicSynth
64
  *A Large-Scale Multilingual Synthetic Speech Dataset for Low-Resource Indian Languages*
 
65
  **πŸ† Outstanding Paper Award, ACL 2025**
66
 
67
  ---
 
70
 
71
  **IndicSynth** is a novel multilingual synthetic speech dataset designed to advance multilingual **audio deepfake detection (ADD)** and **anti-spoofing** research. It covers **12 low-resource Indian languages** and provides both **mimicry** and **diversity** subsets.
72
 
73
+ - 4,000+ hours of synthetic audio
74
+ - 12 languages: Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Odia, Punjabi, Sanskrit, Tamil, Telugu, Urdu
75
+ - Useful for ADD, speaker verification (SV), and bias studies
76
  ---
77
  ## πŸ“‚ Dataset Structure
78
 
 
107
  | vits | TTS | Yes | No |
108
  | freevc24 | VC | No | No |
109
 
110
+ - **Mimicry subset**: For anti-spoofing research
111
+ - **Diversity subset**: Contains diverse set of realistic synthetic voices for multilingual audio deepfake detection research
112
  For more details, please see the Table 1 and Section 3 of our paper: https://aclanthology.org/2025.acl-long.1070.pdf
113
  ---
114
 
 
162
  For questions or feedback, please feel free to reach out at divyas@iiitd.ac.in.
163
 
164
  ## πŸ™ Acknowledgments
165
+ - 🌍 ACL Diversity & Inclusion Subsidy for enabling in-person presentation at ACL 2025
166
+ - 🀝 HuggingFace for dataset hosting support
 
167