CodecFake+: A Large-Scale Neural Audio Codec-Based Deepfake Speech Dataset

A large-scale dataset for deepfake speech detection, created by the CodecFake organization and released in 2025. It includes the CoRS and CoSG subsets, providing audio samples and corresponding protocol and label files for research in synthetic audio generation and detection.

Use Cases

• Training deepfake detection models based on the provided labeled audio samples.
• Benchmarking neural audio codec performance using the CoRS and CoSG subsets.
• Researching synthetic speech artifacts and generation methods based on the dataset's deepfake samples.

Strengths

• Large-scale dataset specifically for neural audio codec-based deepfakes.
• Includes two distinct subsets (CoRS and CoSG) with accompanying label files.
• Public audio samples were released for preview in September 2025.

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.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

CodecFake organization via Hugging Face.

Collection Method

Likely generated using neural audio codec models to create synthetic speech samples.

Time Range

Dataset released in 2025.

Freshness

Last updated 2026-01-20 08:05:46.

Geography

null