According to @AIatMeta, Meta has open-sourced the Perception Encoder Audiovisual (PE-AV), a powerful AI engine underlying SAM Audio’s state-of-the-art audio separation technology (source: @AIatMeta, Dec 18, 2025). PE-AV is built upon the earlier Perception Encoder model and uniquely integrates audio with visual perception, setting new benchmarks in audio and video analysis tasks. The model's native multimodal capabilities enable enhanced sound detection and improved scene understanding, offering significant potential for practical AI applications such as audio forensics, video content analysis, and accessibility solutions. By releasing the code and research paper, Meta is fostering innovation in multimodal AI, opening business opportunities for startups and enterprises aiming to leverage advanced audio-visual machine learning models in commercial products (source: https://go.meta.me/e541b6, https://go.meta.me/7fbef0).