Resemble AI has launched its next-generation deepfake audio detection model, Detect-2B, with an accuracy of 94% and the ability to detect audio in more than 30 languages in 200 milliseconds. Detect-2B uses a series of pre-trained sub-models and fine-tuning technology to identify unique "too clean" features in AI-generated audio, thereby efficiently and accurately identifying deep fake audio and effectively combating AI audio fraud. The model's architecture is based on Mamba-SSM, which has the advantages of fast training and low resource deployment, and has been tested in multiple languages and unknown speakers with excellent performance. Detect-2B will be available via API for easy integration into various applications.
The Detect-2B released by voice cloning company Resemble AI marks significant progress in deepfake detection technology. Its high accuracy, multi-language support and efficient detection speed provide a powerful tool for combating AI audio fraud. In the future, with the continuous development of technology, it is believed that models like Detect-2B will play an increasingly important role in maintaining network security and information authenticity.