341,968 audio recordings across Logical Access and Physical Access categories for the Third Automatic Speaker Verification Spoofing and Countermeasures Challenge. It includes genuine human speech and various spoofing attacks, such as synthetic speech generation and physical replay attempts, to support detection algorithm development.
Use Cases
- Train binary classifiers to distinguish between 'bonafide' and 'spoof' audio using the key column
- Evaluate the performance of speaker verification systems against synthetic speech using the LA subset
- Analyze detection performance across different spoofing algorithms using the system_id feature
Strengths
- Contains 341,968 audio files across the Logical Access (LA) and Physical Access (PA) subsets
- Includes 'bonafide' and 'spoof' labels for binary classification tasks
- Features metadata for 19 different spoofing attack types (A01-A19)