SAM Audio: Separating a Voice from a Noisy Video

SUMMARY

In 30 seconds

Berman demonstrates Meta’s SAM Audio using a short restaurant video. He uses a text prompt to isolate a voice and switches between the isolated and remaining audio; the video is a sponsored demonstration, not a systematic evaluation.

1Audio separationBerman introduces SAM Audio and discloses his partnership with Meta.
2Select a voiceHe enters a prompt to isolate the speaker from restaurant noise.
3Listen to the remainderHe switches off the isolated voice and plays the other sounds.

Takeaways

The demonstration uses a text prompt to select a sound.
The interface provides isolated audio and the remaining sound as separate tracks.
One example does not establish performance on every recording.

TopicsGenerative media Multimodal AI

FULL TEXT

Read the edited text by chapter

Show transcript

Transcript only

SAM Audio demonstration

Berman introduces the sponsored example and its source video.

I demonstrate Meta’s SAM Audio, which can isolate a selected sound from an audio or video recording using a prompt. I am partnering with Meta on this video; the short demonstration is an example rather than an independent benchmark.

The source clip shows a woman speaking on the phone in a noisy restaurant. After playing the original, I enter “voice” and ask the tool to isolate it. The interface produces separate tracks, and I play the isolated speech.

Separate and play the tracks

He isolates the voice and then listens to the remaining audio.

SLIDE 1 / 1Separated audio tracksThe interface shows the original, isolated and remaining sound tracks.

I then mute the isolated voice and play the remaining audio. The example illustrates how a creator can listen to the target and the background separately; it does not establish how well every recording or sound source will separate.