Resemble AI has released its next-generation deepfake detection model, Detect-2B, which is 94% accurate in detecting AI-generated audio. Detect-2B leverages pre-trained sub-models and fine-tuning techniques to detect by identifying unique “too clean” artifacts in AI-generated audio. The model architecture is based on Mamba-SSM and can adapt to different variables and audio qualities, and supports detection of multiple languages. This move is crucial to combating deepfakes and maintaining information authenticity and public trust, especially in the upcoming U.S. presidential election.
Detect-2B launched by Resemble AI is provided through API, making it easy to integrate into various applications. While other companies are developing similar technologies, Resemble AI is committed to continuously improving Detect-2B and plans to conduct further research in aspects such as representation learning, model architecture, and data expansion to meet the challenges of evolving deepfakes technology. This reflects the company’s responsibility and responsibility in maintaining information security and resisting deepfakes threats.