New Delhi: Meta announced SAM Audio; the new model of AI allows users to isolate any sound in complicated audio mixtures. It expands on the Segment Anything series of the company that began with image segmentation. At this point, it has the same power applied to audio. Vocals, instruments or background sound can be separated by users with a few clicks.
The application operates with audio files and videos. It helps in real-life applications such as in cleaning up podcasts or remixing songs. Meta refers to it as the original audio separation model. It deals with various tasks more efficiently compared to the current tools.
How SAM Audio Works
SAM Audio applies three prompt types. They can be used together by the users to give accurate results.
Text prompting will allow you to explain the sound. It can be removed with the type “guitar solo” or “dog barking”.
Visual cues are used on videos. Click on an object or person. The model separates the sound that it produces.
Span prompts are used to mark time. Point out the position of sound. The AI scans and isolates it throughout the file.
The techniques render editing to be intuitive. No advanced skills needed.
Applications and impact
SAM audio dominates artistic circles. Single instruments can be drawn out by musicians in band recordings. The podcasters are able to eliminate the noise that is not intended easily. There is dialogue cleaning in location shoots by filmmakers.
It also aids accessibility. Meta also collaborates with hearing aid manufacturers in order to enhance sound. Studies can be conducted to examine particular audio.
Its model ranks higher than competitors. Meta released it open-source. Developers, download it now.
Availability and limitations
Use SAM Audio in the Meta Playground of Segment Anything. Upload files or use samples. Complete examples are available at GitHub and Hugging Face.
It has limits. The device has a weakness in the case of very similar sounds, as in a choir. It requires prompts and is not yet able to accept audio examples. Meta plans improvements.