Tandem Detection Cost Function (T-DCF)
Definition
The primary evaluation metric in ASVspoof from 2019 onward. It measures the cost of errors when a countermeasure is integrated with an automatic speaker verification system, weighting false accepts and false rejects by their operational costs.
- First used
- ASVspoof challenge, from 2019 onward
- Measures
- Cost of errors in an ASV system paired with a spoofing countermeasure
- Weights
- False accepts and false rejects by operational cost
- Field
- Voice conversion and cloning detection evaluation
Common questions
Why did ASVspoof move from a standalone countermeasure metric to the t-DCF?+
A countermeasure evaluated on its own, such as by equal error rate for spoof detection, does not capture how it actually affects the combined system's real-world performance once integrated with a speaker verification system, so t-DCF was introduced to score the countermeasure and ASV system together as they would operate in deployment.
What does a low t-DCF score actually indicate for a deployed voice authentication system?+
A low t-DCF indicates that when the countermeasure is added in front of the speaker verification system, the combined cost of wrongly accepting a spoofed or impostor voice and wrongly rejecting a genuine speaker is minimised according to the assumed operational cost weighting, not just that the countermeasure alone detects spoofing well in isolation.
Can a countermeasure with excellent standalone spoof-detection accuracy still produce a poor t-DCF?+
Yes; if the countermeasure introduces a high rate of false rejections for genuine speakers when combined with the ASV system, or if it interacts poorly with the specific verification system's decision threshold, the tandem cost can be worse than its standalone detection accuracy would suggest.
Related terms
- Anti-Spoofing Countermeasure (CM)
- A classifier, also called a CM system, trained to output a score indicating the probability that a given audio segment is genuine...
- ASVspoof
- A recurring evaluation campaign and dataset series that benchmarks anti-spoofing countermeasures against corpora of genuine and spoofed utterances. Editions in 2015, 2017,...
- Equal Error Rate (EER)
- The point on a classifier's detection error tradeoff curve where the false accept rate equals the false reject rate. Lower EER indicates...
- Neural TTS Cloning
- A text-to-speech system that adapts to a target speaker using a short enrollment recording, generating new utterances in that speaker's voice from...
- Voice Conversion
- A signal-processing or deep-learning technique that transforms the vocal characteristics of a source speaker's utterance to match a target speaker, while preserving...