Boris FX CrumplePop 2026.5 [WiN-MAC]

• Product: CrumplePop
• Developer: Boris FX
• Version: 2026.5.18
• Format: VST3, AU, AAX
• Requirements: Windows 10 or later, macOS 12.0 or later
• Source: borisfx.com/products/crumplepop

CrumplePop 2026.5 bundles an AI demixing application, ARA2 session plugins, and one-knob VST/AU/AAX inserts under one license. SoundApp separates mixed audio into voice, drum, bass, and dialogue stems using proprietary models run locally on GPU. The ARA build reads a clip’s full context before separating it inside Pro Tools, Sequoia, and Reaper sessions, while the plugin versions handle single-pass denoise and echo removal inside Premiere and Resolve timelines. A podcast editor drags a phone-recorded interview into SoundApp and pulls hum out before the dialogue reaches a compressor.

Key Takeaway

CrumplePop activates when a mixed file needs breaking into stems, or a raw recording needs denoising without reopening the original session. It complements restoration work rather than replacing multitrack mixing. The ARA engine runs only inside ARA2 hosts; Premiere, Resolve, and Avid users get the simpler real-time plugins instead. A mixer already holding clean stems has no use for it.

Four Stems From One Mixdown

Music 4 Stems pulls a finished mixdown apart into voice, drums, bass, and an “other” bucket for guitar, synth, or pads. The separation runs off proprietary trained models rather than a frequency-band filter, so a vocal line buried under a wide guitar tone still comes out on its own track. A drummer’s overhead bleed into a vocal mic gets resolved the same way — by model inference, not EQ carving.

Voice 2 Stems runs the same engine on two outputs instead of four, isolating dialogue or lead vocal against everything else in the file. A remixer building a karaoke bed from a finished master gets an instrumental track without needing session files that may not exist anymore. Cinema 4 Stems repeats the split for picture-locked audio, breaking a rendered scene into dialogue, music, effects, and ambience.

The separation works from the exported file alone, so artifacts already baked into the mix — clipping, heavy compression, a vocal already ducked under a synth pad — carry into the isolated stem instead of disappearing with it. A stem pulled from a badly clipped source still carries that clipping.

Reads the Whole File First

The ARA build of SoundApp analyzes an entire clip’s waveform before separating or denoising it, instead of processing in the short buffers a real-time plugin is limited to. That full-file pass changes what the separation can resolve — a vocal that drops out for two bars and returns has continuous context on either side, rather than each buffer being evaluated in isolation.

It runs inside Pro Tools, Sequoia, Samplitude, Reaper, and Cubase — any ARA2 host. Premiere, Resolve, Final Cut, and Avid sessions don’t get this engine; they get the separate one-knob VST/AU/AAX plugins instead.

A processed-audio cache holds the result once a model runs on a clip, so switching models or duplicating a processed instance across a project doesn’t retrigger analysis. Model load times dropped in the 2026.5 release, cutting the pause between selecting a model and hearing the first processed frame.

Stacked Plugins, Not One Fix-All

AudioDenoise, EchoRemover, WindRemover, TrafficRemover, ClipRemover, PopRemover, RustleRemover, and Levelmatic run as standard VST3, AU, or AAX inserts inside Premiere, Resolve, Final Cut, Avid Media Composer, and Vegas timelines. Each targets one noise type — traffic, wind, plosives, a clipped peak — rather than one plugin attempting all of them.

Processing runs in real time on the timeline, without the full-file pass the ARA build performs, so playback starts immediately when the plugin loads. A podcast episode with wind noise on one segment and lav rustle on another gets two separate plugins stacked on the clip rather than one model switching context mid-file.

Real-time processing means each instance evaluates a rolling buffer rather than the whole clip at once, so a noise print that changes character partway through a take gets a less consistent result than the same material would through the ARA engine’s full-file pass.

No Upload Queue, No Credits

SoundApp’s models run on the local GPU rather than a remote server, so a file never leaves the machine during separation or denoising, and there’s no per-export credit balance to track. Processing on a longer file starts as soon as the playhead lands on it — seek and go — instead of waiting for the full clip to finish rendering somewhere else first.

A field recordist working on an unreleased interview keeps the audio off any third-party server for the entire cleanup pass. Export volume isn’t metered against a subscription tier the way cloud-based demixing services typically gate it.

Local GPU processing means output speed depends on the host machine’s hardware — SoundApp calls for 16GB RAM and meaningfully more disk space than the 8GB / 450MB footprint the plugin-only install needs. A laptop without a strong GPU path runs SoundApp slower than the lighter plugin versions, which don’t carry the same dependency.

Presence Correction, After the Cleanup Pass

Voice Enhance runs as a single-control plugin that reworks tone and presence on a vocal track without a manual EQ or de-esser chain. It’s built to run after another CrumplePop 2026.5 tool in the chain — a rustle or wind removal pass first — rather than to replace a full restoration signal chain on its own.

A laptop-mic recording or a phone-shot interview gets brought toward the tonal balance of a treated-room vocal, adjusted by the model’s own thresholds rather than manual gain staging. Zoom-call audio and distant cellphone recordings are named use cases specifically because those sources carry the most inconsistent frequency response session to session.

Voice Enhance is built for spoken dialogue and vocal tone correction specifically — it isn’t a noise-type remover on its own, so a clip with heavy traffic or wind still needs one of the dedicated removal plugins run first, per the documented signal-chain order.

No ARA Demixing Inside Premiere or Resolve

A video editor cutting inside Resolve reaches for EchoRemover or WindRemover on the timeline directly, processing a rolling buffer instead of the full clip SoundApp’s ARA engine reads before it splits anything.

FAQs