Datasets:
Tasks:
Audio Classification
Sub-tasks:
multi-class-classification
Languages:
English
Size:
10K - 100K
Upload binary_balanced/README.md with huggingface_hub
Browse files- binary_balanced/README.md +35 -0
binary_balanced/README.md
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# amine-maazizi/vc-detection-10k (binary_balanced)
|
| 2 |
+
|
| 3 |
+
## Dataset Purpose
|
| 4 |
+
This dataset is built for student-scale anti-spoofing and deepfake audio detection experiments focused on voice conversion (VC).
|
| 5 |
+
|
| 6 |
+
## Source Corpus
|
| 7 |
+
- Real speech source: VCTK subset extracted from local tar shards.
|
| 8 |
+
|
| 9 |
+
## Fake Generators
|
| 10 |
+
- DDDM-VC
|
| 11 |
+
- Diff-HierVC
|
| 12 |
+
|
| 13 |
+
## Label Schema
|
| 14 |
+
- Variant: binary_balanced
|
| 15 |
+
- real
|
| 16 |
+
- fake
|
| 17 |
+
- Variant: multiclass_by_generator
|
| 18 |
+
- real
|
| 19 |
+
- fake_dddm_vc
|
| 20 |
+
- fake_diff_hiervc
|
| 21 |
+
|
| 22 |
+
## Split Methodology
|
| 23 |
+
Speaker-disjoint train/val/test splits are enforced at subset construction time.
|
| 24 |
+
|
| 25 |
+
## Limitations
|
| 26 |
+
- Generated from two VC systems only.
|
| 27 |
+
- Objective labels are pipeline-derived and may include conversion failures if not filtered.
|
| 28 |
+
- Short clips (roughly 1-5s) and 16 kHz mono assumptions.
|
| 29 |
+
|
| 30 |
+
## Storage Assumptions
|
| 31 |
+
- WAV audio, 16 kHz, mono, int16.
|
| 32 |
+
- Metadata includes duration and sample rate.
|
| 33 |
+
|
| 34 |
+
## Intended Use
|
| 35 |
+
Academic prototyping and benchmarking for VC-focused spoof detection tasks.
|