Meta, 음성 분리 AI 모델 SAM Audio 공개
무슨 발표인가
- 텍스트·시각·시간 범위 프롬프트로 복합 오디오에서 음성 격리
- 음악·팟캐스트·영상 편집·접근성·과학 연구 등 다양 분야 적용
- Segment Anything 컬렉션의 최신 모델
원문 (영어)
Today, we’re introducing SAM Audio, a state-of-the-art AI model that enables you to segment sound. Imagine recording a video of your favorite band and isolating the guitar or vocals with a single click, using text prompts to filter traffic noise from a video filmed outside, or removing the sound of a dog barking from your entire podcast recording.
SAM Audio, the latest addition to our Segment Anything collection , transforms audio processing by making it easy to isolate any sound from complex audio mixtures using text, visual, and time span prompts. This intuitive approach mirrors how people naturally engage with sound, making professional-grade audio separation more accessible and easier than ever before.
SAM Audio has the potential to transform audio and video editing and drive innovation in areas like music, podcasting, television, film, scientific research, accessibility, and more. Until now, audio segmentation and editing has been a fragmented space, with a variety of tools designed for single-purpose use cases.
As a unified model, SAM Audio is the first to support use cases that match how people naturally think about audio, and achieves cutting-edge performance across diverse, real-world scenarios. SAM Audio supports three kinds of prompts: Text prompting : Type “dog barking” or “singing voice” to extract specific sounds.
https://about.fb.com/wp-content/uploads/2025/12/01_Text-Prompts.mp4 Visual prompting : Click on the person or object in the video that’s making a sound to isolate their audio. https://about.fb.com/wp-content/uploads/2025/12/02_Visual-Prompts.mp4 Span prompting : An industry first, this method lets you mark time segments where target audio occurs.
https://about.fb.com/wp-content/uploads/2025/12/03_Span-Prompts.
원문: Meta Newsroom — "Our New SAM Audio Model Transforms Audio Editing" (2025-12-16) 공식 원문: https://about.fb.com/news/2025/12/our-new-sam-audio-model-transforms-audio-editing/
A small team that reads too much internet so you don't have to.
