- Author
- Posted by megaman821, with replies from many commenters
- Published
- 16 December 2025
- Kind
- Hacker News discussion thread, 223 points
- Original
- news.ycombinator.com/item?id=46292266
- Subject
- Music tools and instruments
What it is
The thread was posted by megaman821 on Hacker News. It links Meta's SAM Audio page at ai.meta.com/samaudio. Commenters try the demo, discuss uses and argue about music technology and rights.
What it says
- The thread links Meta's SAM Audio page. One early comment praises the demo for separating music, voice and background noise.
- Testers report mixed results. Footsteps, air conditioning and voices worked, but "train noises", "background noise" and "clatter" mostly failed.
- One commenter found the output near identical to the original when isolating speech or music.
- One commenter relays GitHub issue reports on hardware. One user managed only 20s clips on an RTX 5090.
- One commenter argues separation is underconstrained, so a model may invent parts.
- Commenters want it for stems, learning songs, single-microphone recording and removing laugh tracks.
What evidence it gives
The evidence is anecdotal. Commenters describe their own prompts and results, and one relays reports from a GitHub issue. Nobody in the thread posts measurements.
Interests and limits
Commenters are self-selected and mostly tried a web demo. Claims about hardware needs come secondhand from an issue thread that was not opened. Meta's page itself was not read. Several comments are off topic or flagged.