We study synthetic speech detection across diverse voice generation systems, recording environments, and real-world conditions.
Explore Our Research Findings →Research areas focused on improving synthetic speech detection under practical conditions.
Can detection systems identify synthetic speech generated by models they have never encountered before?
How does performance change under compression, noise, and low-quality recordings?
How can detection systems remain useful outside controlled laboratory environments?
Monitoring advances in voice cloning and synthetic media capabilities.
We evaluate detection performance across commercial systems, open-source models, benchmark datasets, and real-world recordings.
Engineering decisions are based on measurable evaluation results.
Models are tested regularly against new datasets and recording conditions.
Performance is measured on separate evaluation data to better estimate generalization.
Testing emphasizes realistic recording conditions rather than ideal laboratory audio.
Testing includes a range of speech generation methods and recording conditions to reduce overfitting.
| Testing Array | Evaluation Status |
|---|---|
| Commercial Voice Models | Ongoing |
| Open-Source Models | Ongoing |
| Benchmark Datasets | Ongoing |
| Meeting & Recorded Audio | Ongoing |
Current areas of development.
If you are coordinating an evaluation framework, exploring acoustic benchmarks, or have technical inquiries, reach out to our team.