Can detect audio in more than 30 languages in just 200 milliseconds
Detect-2B uses a series of pre-trained sub-models and fine-tuning techniques specifically used to inspect audio clips and determine whether they were generated by AI.
“Based on our original Detect model, Detect-2B has made significant improvements in model architecture, training data and overall performance. The model was evaluated on a large dataset of real and fake audio clips and demonstrated impressive performance.” Resemble said in an official blog post.
According to Resemble, the sub-model of Detect-2B consists of a frozen audio representation model with a key layer insertion adaptation module. These adaptation modules focus on identifying the nuances between real audio and fake audio-that is, the sound traces inadvertently left behind during the recording. Most AI-generated audio clips sound “too perfect.” Detect-2B is able to predict AI-made components in audio without having to retrain the model every time a new clip is heard. These sub-models have also been fully trained on large data sets.
Detect-2B will summarize its predicted scores and compare them to “carefully adjusted thresholds” to determine the authenticity of the recording. Resemble said that their unique design allows Detect-2B to train quickly and without requiring a lot of computing power when deployed.
Identifying deep forgeries has become particularly important
As the 2024 U.S. presidential election approaches, recognizing AI-generated sounds or videos becomes increasingly important. AI voices can increase the risk of misleading voters and spreading misinformation. Whether it’s faking a politician’s voice, impersonating a celebrity in a song, or simply using AI to state something, concerns about AI’s deep counterfeiting have eroded the public’s trust in brands.
Tools like Detect-2B can largely help identify and prove the counterfeiting of these deeply forged content before it enters the public eye. Of course, Resemble is not the only company dedicated to testing AI clones. For example, McAfee launched the Project Mockingbird project in January to detect AI audio. Meta is developing a way to add watermarks to AI-generated audio.
“But our work is far from over. As generative AI capabilities continue to increase, our detection capabilities must also be improved simultaneously. We have planned several exciting research directions to further optimize Detect-2B, focusing on areas such as representational learning, advanced model architecture and data expansion.” Resemble said.
If you want to learn more, you can click on the link below the video.
Thank you for watching this video. If you like it, please subscribe and like it. thank
Original text:https://venturebeat.com/ai/resemble-ais-next-generation-ai-audio-detection-model-detect-2b-is-94-accurate/
Oil tubing: