SAM Audio is live · free · last verified 24 Oct 2026 · viability monitor
SAM Audio
SAM Audio separates any sound from a mix using text, visual, or span prompts — Meta's open-source audio separation model.
Judge SAM Audio as a research release, not a SaaS product. Span prompting is the genuine news here, and one model spanning general sounds, music, and speech beats buying a speech-only or music-only separator. If you want a click-and-go app with a support contract, this isn't your tool — iZotope RX or Acon Digital are. With a GPU and a specific promptable-separation problem, download it.
Verdict generatedVerified 6d ago · liveness 61/100 · cite: rightaichoice.com/tools/sam-audio
- Audio engineers prototyping prompt-based source isolation
- Video editors isolating dialogue or sound effects from footage
- Multimedia and audio researchers
- Developers comfortable running open model weights on their own hardware
- Production apps needing a managed endpoint with an uptime commitment
- Teams that need a support contract when output quality regresses
- Buyers who want separation as a point-and-click application rather than a model to run
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip SAM Audio if you need separation as a managed endpoint with an uptime commitment and a support contract — this is a downloadable research model you run on your own GPU, not a hosted service.
Running the downloaded weights yourself means paying for your own GPU compute, and inference is compute-heavy enough that the page recommends a GPU for fast results.
SAM Audio is released as a free open-source model rather than a paid product, so the comparison is not tier pricing but cost of ownership: you supply the GPU and engineering time. Against commercial separation suites such as iZotope RX or Acon Digital, you trade a licence fee for infrastructure and integration work. Against other open models, the bundled open evaluation set and judge model are the differentiator.
In short
SAM Audio — SAM Audio separates any sound from a mix using text, visual, or span prompts — Meta's open-source audio separation model. Best for Audio engineers prototyping prompt-based source isolation, Video editors isolating dialogue or sound effects from footage, Multimedia and audio researchers. Free to use.
What is SAM Audio?
SAM Audio — SAM Audio separates any sound from a mix using text, visual, or span prompts — Meta's open-source audio separation model. Best for Audio engineers prototyping prompt-based source isolation, Video editors isolating dialogue or sound effects from footage, Multimedia and audio researchers. Free to use.
SAM Audio is an open-source audio separation model from Meta that pulls a target sound out of any audio or audiovisual mixture. The pitch is promptability: describe what you want in text ("dog barking"), click on the object in a video that produces the sound you're after, or select the timespan containing it. Meta says SAM Audio is the first model to introduce span prompting, and the model unifies general sounds, music, and speech under those same three prompt modalities instead of one specialist model per domain. The output convention is unusually useful. Architecture-wise it's a generative separation model built on a flow-matching Diffusion Transformer operating in a DAC-VAE latent space, and it extracts both the target stem and the residual stem — the thing you asked for plus the leftover mix — enabling joint high-quality generation of both. Meta shipped supporting pieces alongside it: Perception Encoder Audio Video (PE-AV), an open-source audiovisual encoder that brings audio capability to the Perception Encoder family, plus a first-of-its-kind open evaluation set for prompted audio separation and a judge model Meta describes as highly correlated with human subjective evaluation. Pretrained weights and inference code are downloadable, and a hosted playground lets you try separation before installing anything. The audience is audio engineers, video editors, multimedia researchers, and developers wiring separation into a larger pipeline. Meta frames it as a research release, and the quoted reference partners — 2gether-International and hearing-aid maker Starkey — use it as an open model inside their own R&D rather than as a subscription service. For a managed endpoint with an uptime commitment, look at commercial separation suites instead.
Key features
- Text-prompted audio separation
- Visual prompting by clicking the sound's source in video
- Span prompting — select the timespan containing the target audio
- Three unified prompt modalities: text, visual, timespan
- Separates both target and residual stems
- Unified separation across general sounds, music, and speech
- Flow-matching Diffusion Transformer architecture
- Operates in a DAC-VAE latent space
- Audiovisual perception via the open-source PE-AV encoder
- First-of-its-kind open evaluation set for prompted audio separation
- Judge model correlated with human subjective evaluation
- Pretrained model weights available to download
- Public playground to try separation before installing
- Open-source release from Meta
Behind the verdict
The first question worth asking about SAM Audio is who's actually going to run it. The weights are downloadable and the code ships alongside, so the honest answer is anyone with a GPU and a Python environment. If that isn't you, the playground is a fine way to see whether prompt-based separation fits your workflow before you commit to the setup — but it's a look, not a deployment plan. We'd reach for SAM Audio when the target sound is awkward to describe to a classic source-separation model that only splits vocals and instruments. Click-on-the-source visual prompting is the feature that earns its keep for video editors: scrub to the shot, click the object, get the sound. Span prompting is Meta's claimed first, and it's the one that maps most naturally to the audio work people actually do — you know when the car horn happens, you just want that second isolated. The target-plus-residual output is the smart part. Most separators hand you one stem and leave you to figure out what's left. Getting both at once means you keep the rest of the mix intact instead of reconstructing it, which matters when you're prototyping rather than finishing a master. Where it bites: this is research code. There's no uptime SLA, no support line when output quality regresses on your material, and no point-and-click application layer. The reference partners on the page — 2gether-International and Starkey — use it as an open model inside their own R&D, which tells you the intended deployment is a team with ML staff, not a solo editor looking for a plugin. The closest alternative depends on your axis. For promptable open research, it's the model to beat on breadth. For commercial, click-and-go isolation with someone to call, the audio-restoration suites are the category you should be shopping
Models under the hood
as of 2026-09-14
Limitations
- Meta presents SAM Audio as a research release — a model you download or try in a playground, not a managed service.
- Inference is compute-heavy and a GPU is recommended for fast results.
- In the released setup, input is single-channel.
- Highly overlapping sources are hard for any separation model, and Meta's own framing centers on evaluation rigor (an open eval set plus a judge model) rather than a universal quality guarantee.
- Licensing and commercial-use terms for the weights are something your legal team should review before you build a product on them.
as of 2026-10-13
Researching SAM Audio? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is SAM Audio still active?
SAM Audio is live · free · last verified 24 Oct 2026 · viability monitor.
Viability score
How well maintained and how widely used is SAM Audio? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →What do users say about SAM Audio?
What people actually say about SAM Audio — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
Researched37 mentions across 4 sources (Hacker News, YouTube, Product Hunt, Lemmy) · researched Oct 20, 2026.
Weighted by the 45 posts each of 4 sources contributed.
- +Reported ~17% net win over Demucs on MUSDB18 makes the quality claim credible
- +Three unified prompt modes (text, visual, span) feel genuinely novel and useful
- +Span prompting is a first-of-its-kind capability that nails tricky time-localized targets
- +Open weights and code mean no vendor lock-in and real auditability
- +Extracts both target and residual stems in one pass, saving post-production work
- −Official implementation often needs 30GB+ VRAM, strains even workstation-class cards
- −Multiple users on gaming-spec hardware could not get it running at all
- −No managed API, SLA, or hosted production endpoint from Meta
- −Benchmark chart ambiguity leaves users unsure how big the quality win is
- −Research-release posture means limited docs, support, and long-term maintenance
- • GPU compute — 30GB+ VRAM on the official repo, or cloud GPU rental for those without hardware
- • Engineering time to package, serve, and maintain your own inference stack
- • Community wrapper dependencies (fal, Playwright, Modal-hosted variants) add third-party costs and reliability concerns
How we verified this
Every listed fact about SAM Audio— pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 24 Oct 2026. Pages mix model-assisted synthesis with human review; see how we verify.
Verification history
We have re-verified SAM Audio 9 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed SSL certificate
- — changed SSL certificate
- — changed who it suits, summary, pricing tiers, description, our verdict, who should skip it, features
- — changed SSL certificate
- — changed our verdict, description, pricing fit, setup time, who should skip it, hidden costs, limitations, use cases, switching in, switching out
- — changed SSL certificate
Showing the 6 most recent of 9 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does SAM Audio cost in 2026?
SAM Audio: Free to use. Pricing last verified 24 Oct 2026.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where SAM Audio's pricing actually pencils out — and where peers do it cheaper.
SAM Audio is released as a free open-source model rather than a paid product, so the comparison is not tier pricing but cost of ownership: you supply the GPU and engineering time. Against commercial separation suites such as iZotope RX or Acon Digital, you trade a licence fee for infrastructure and integration work. Against other open models, the bundled open evaluation set and judge model are the differentiator.
Has SAM Audio's pricing changed?
No plan price for SAM Audio has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick SAM Audio?
Pick SAM Audio if you are Audio engineers prototyping prompt-based source isolation, Video editors isolating dialogue or sound effects from footage, Multimedia and audio researchers, Developers comfortable running open model weights on their own hardware.
Look elsewhere if you are Production apps needing a managed endpoint with an uptime commitment, Teams that need a support contract when output quality regresses, Buyers who want separation as a point-and-click application rather than a model to run.
Skip it if Skip SAM Audio if you need separation as a managed endpoint with an uptime commitment and a support contract — this is a downloadable research model you run on your own GPU, not a hosted service.
GeneratedReal-world workflow fit
Concrete scenarios for the personas SAM Audio actually fits — and what changes day-one when you adopt it.
You have a street interview with traffic and a busker in the background. You import the clip, use visual prompting to click the busker in frame and pull that audio into a residual stem, then take the target dialogue stem clean.
Outcome: You keep the music bed and room tone as a separate residual file instead of losing the scene's atmosphere to a blunt noise gate.
You download the pretrained weights and inference code, run them against Meta's first-of-its-kind open evaluation set for prompted audio separation, and compare scores using the released judge model.
Outcome: You get a reproducible baseline for your own separation approach without building an evaluation harness from scratch.
You wire SAM Audio's three prompt modalities — text, visual click, span — into an internal tool, testing span prompting on a long field recording where describing the target sound in words is unreliable.
Outcome: You learn which prompt modality your users actually reach for before committing engineering time to a product feature.
Use cases
- Isolate one conversation from a crowded restaurant recording with a text prompt naming the speech you want.
- Pull a guitar track out of a song by clicking the guitar in the video.
- Extract a dog barking sound effect from a street recording using a span prompt on the timeline.
- Keep the music bed and room tone while cleaning dialogue, by taking both the target and residual stems.
- Analyze wildlife field recordings by prompting for a specific bird call.
- Prototype promptable separation inside a research pipeline using the downloadable weights.
- Build an audiovisual separation feature on top of the open PE-AV encoder.
- Benchmark your own separation approach against Meta's open evaluation set and judge model.
Setup time & first value
How long it actually takes to get something useful out of SAM Audio — broken out by persona, not the marketing-page minute.
If you take the playground route, first useful separation is minutes — no install. If you download the weights, budget an afternoon: Python environment, GPU, and getting inference code running. Teams evaluating vs. a managed suite should budget a few days including licence review.
Switching to or from SAM Audio
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From a commercial separation suite like iZotope RX: download the weights and re-run your stems through SAM Audio to compare prompted separation against your existing spectral-editing workflow.
- →From a specialist vocal or stem remover: replace the fixed domain filter with text, visual, or span prompts so the same model handles general sounds, music, and speech.
- →From a hand-rolled separator built on an audiovisual encoder: swap in Meta's open-source PE-AV encoder, which brings audio capability to the Perception Encoder family.
- ↗To a managed separation API: if you need an endpoint with an uptime commitment, move the pipeline to a hosted vendor and keep SAM Audio only for offline batch work.
- ↗To a commercial audio repair suite like iZotope RX or Acon Digital: if your team wants point-and-click separation with vendor support, hand the stems to a tool built for editors rather than researchers.
- ↗To your own trained separation model: use Meta's open evaluation set and judge model to benchmark an in-house model, then retire SAM Audio from the pipeline once yours wins.
Integrations
Resources & guides
Tutorials & Learning
Official links
What are the best SAM Audio alternatives?
The closest alternatives to SAM Audio we list are maestro SFX by beatoven.ai, Cleanvoice and Dolby On. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| maestro SFX by beatoven.ai | Freemium | Pricing not published | — |
| Cleanvoice | Freemium | Pricing not published | — |
| Dolby On | Free | Free | — |
Alternatives to SAM Audio
View all
maestro SFX by beatoven.ai
AI sound effect generator that turns text prompts into production-ready, royalty-free audio in seconds.
Cleanvoice
Cleanvoice is an AI podcast editor that strips filler words, dead air, mouth sounds and background noise from audio and video in minutes.
Tools that pair well with SAM Audio
Common stack mates teams adopt alongside SAM Audio, with the specific reason each pairing earns its keep.
maestro SFX by beatoven.ai
AI sound effect generator that turns text prompts into production-ready, royalty-free audio in seconds.
Cleanvoice
Cleanvoice is an AI podcast editor that strips filler words, dead air, mouth sounds and background noise from audio and video in minutes.
Dolby On
Free Dolby mobile app that records audio, video, and livestreams with one-tap sound enhancement
SAM Audio vs its closest rival
SAM Audio's closest head-to-head is with Splice — If you need a vast library of royalty-free samples and rent-to-own plugins, Splice's monthly plans (starting at $4.99) are unmatched. But if you require precise, prompt-based audio separation for speech or effects, SAM Audio's free, open-source model is a powerful, cost-free alternative. Choose based on your core workflow: sample sourcing vs. source isolation. Verdict generated
Head-to-head comparisons
Sam Audio vs Splice
If you need a vast library of royalty-free samples and rent-to-own plugins, Splice's monthly plans (starting at $4.99) are unmatched. But if you require precise, prompt-based audio separation for speech or effects, SAM Audio's free, open-source model is a powerful, cost-free alternative. Choose based on your core workflow: sample sourcing vs. source isolation.
GeneratedSam Audio vs Landr Mastering
LANDR and SAM Audio serve completely different purposes: LANDR is for mastering finished tracks, SAM Audio for separating mixed audio. Pick LANDR if you need a quick, AI-powered master with reference matching and album cohesion (starting at $10/track). Choose SAM Audio if you need to isolate speech, music, or effects from a single audio file — it's free and open-source but requires technical setup.
GeneratedSam Audio vs Storyfile
For museums or legacy projects needing authentic interactive human avatars, StoryFile is the clear choice. For audio professionals who need to separate sounds with simple prompts, SAM Audio is free and state-of-the-art. They solve entirely different problems, so your decision hinges on whether you need conversational video AI or audio source separation.
GeneratedFrequently asked questions
Categories
Used SAM Audio? Help shape our editorial sentiment research.