ASVspoof 2015: Genuine and Spoofed Speech for Speaker Verification Security

106 speakers (45 male, 61 female) contributed genuine speech recordings with minimal channel or background noise. The database includes spoofed speech generated from the genuine data using several different spoofing algorithms. It is partitioned into training, development, and evaluation subsets for use in the ASVspoof 2015 challenge.

Use Cases

• Training machine learning models to detect spoofed speech based on the provided genuine and spoofed audio samples.
• Benchmarking spoofing countermeasure algorithms using the defined training, development, and evaluation partitions.
• Studying the effectiveness of different spoofing algorithms mentioned in the description on speaker verification systems.

Strengths

• Contains genuine speech from 106 distinct speakers.
• Spoofed data is generated using multiple algorithms, providing diverse attack scenarios.
• Structured partitioning into training, development, and evaluation subsets supports standardized benchmarking.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and total dataset size are unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Zhizheng Wu

Collection Method

Genuine speech collected from speakers; spoofed speech generated algorithmically from the genuine data.