SAM segments objects in images and videos, even audio can be separated by prompt: The AI model is freely available.

不同于传统的视觉-语言模型（VLM）通过自回归方式生成 token，VL-JEPA 预测的是目标文本的连续嵌入（embedding）。通过在抽象的表征空间中学习，该模型能够专注于与任务相关的语义，同时忽略表层语言形式的多变性 。

meta公司近日在音频技术领域取得重大进展，正式发布全球首个多模态音频分离模型——SAM Audio。这项创新技术通过模拟人类感知声音的天然方式，实现了对复杂音频的精准解析与交互式提取。用户现在能够像“用眼睛聆听”般，从混合音频或视频中分离出特定目标声音，无论是点击画面中的乐器、输入文字描述声源，还是标记时间片段，均可一键完成操作。 该模型的核心突破在于其自研的感知编码器视听引擎（PE-AV），这 ...

Meta正式推出音频处理领域的重磅突破——SAM Audio，全球首个统一的多模态音频分离模型。它能让用户像“用眼睛听声音”一样，从一段混杂的视频或音频中，一键提取出任意目标声音:点击视频中的吉他手，立刻分离出纯净吉他声;输入“狗吠”，自动过滤掉整段 ...

Meta has released an open-source AI model called SAM Audio that lets users clean up noisy recordings by describing what they want to remove. The tool can isolate voices, music, or background noise ...

Meta 表示 SAM Audio 是一个“最先进的统一模型”，通过使用自然的、多模态的提示，使音频处理变得简单， 能够轻松地从复杂的音频混合中分离出任何声音 —— 无论是通过文本、视觉提示还是时间段标记。这种直观的方法模拟了人们自然与声音互动的方式，使音频分离更加易于使用和实用。

Meta Platforms Inc. is bringing prompt-based editing to the world of sound with a new model called SAM Audio that can segment individual sounds from complex audio recordings.

The final, formatted version of the article will be published soon. Purpose: Risk perception significantly impacts how individuals assess risk, make decisions, and behave. While numerous studies have ...

Meta plans to cut its budget by up to 30% in its Reality Labs metaverse division. It's considering job cuts as part of that move, leaving employees uncertain. The cuts may impact Horizon Worlds and ...

Meta plans to direct its investments to focus on wearables like its augmented reality glasses but does not plan to abandon building the metaverse. By Mike Isaac Reporting from San Francisco. Meta is ...