AI · AI model
SAM-Audio
Segment and isolate specific sounds or voices using multimodal prompts.
SAM-Audio is an audio segmentation model from Meta designed to isolate targeted sounds from audio and video recordings. It supports multimodal prompting, allowing users to isolate audio elements using text descriptions, visual bounding boxes on video frames, or specific time spans.
What it's for
- Isolate a specific speaker or voice from a noisy video recording
- Separate individual musical instruments from an audio track using text prompts
- Extract environmental sounds by selecting visual objects in video footage
Features
- Multimodal prompting via text descriptions, visual bounding boxes, and waveform time spans
- Targeted sound and voice separation from mixed audio
- Integration with video context for object-directed audio extraction
Pricing
Free model weights on Hugging Face under custom research license
Details
- Type
- AI model
- Category
- AI
- Topic
- Audio Separation
- Open source
- No
- Platforms
- Code library, Self-hosted
- For
- Audio engineers, video creators, and AI researchers
- Alternative to
- LALAL.AI, Audacity noise removal
- Tags
- audio, meta, stem separation, multimodal, sound isolation
Why it was saved
- @evolving.ai: Allows creators and sound engineers to separate targeted audio elements and instruments directly from video context.
Related
AI platform to generate, search, and score royalty-free music for videos
AI voice generator, voice cloning, and conversational agent platform
AI platform that generates complete songs with vocals and instrumentation from text.
Web-based AI audio recording, editing, and speech enhancement tool
Open full-duplex speech-to-speech conversational AI model by NVIDIA
Ultra-low-latency streaming text-to-speech and instant voice cloning model.
Updated