Skip to content

Releases: danielraffel/pulp-spectral-lab

Spectral Lab 1.0.2

Choose a tag to compare

@danielraffel danielraffel released this 04 Jul 13:42

Spectral Lab 1.0.2 — stability fix

Fix: no more crash when you add the plugin and then remove it from a track. Tearing the editor down could race the GPU UI's frame callback, which could crash the host on delete. Editor teardown is now serialized on the main thread and the frame callback re-checks that the view is still alive, so add → remove is safe in Logic (AU), REAPER/Cubase/Bitwig (VST3/CLAP), and Standalone.

Audio and UI are otherwise unchanged from 1.0.1.

One notarized installer (Customize → AU / VST3 / CLAP / Standalone). Apple Silicon; run your host natively (not Rosetta).

Spectral Lab 1.0.1

Choose a tag to compare

@shipyard-local shipyard-local released this 28 Jun 09:14

Spectral Lab 1.0.1 — bug fix

Freeze now sustains a smooth cloud instead of a buzzy loop. A coherent freeze repeated exactly every FFT period (~43 ms), which sounded like a tight half-second loop. The per-hop phase wander (Jitter) is now scaled properly and defaults higher, so a freeze decorrelates that repetition and sustains the actual spectral content as a smooth, evolving pad. Lower Jitter toward 0 for a more static, tonal freeze.

One notarized installer (Customize → AU / VST3 / CLAP / Standalone). Apple Silicon; run your host natively (not Rosetta).

Spectral Lab 1.0.0

Choose a tag to compare

@shipyard-local shipyard-local released this 27 Jun 23:24

Spectral Lab 1.0.0 — GPU spectral freeze + morph

Capture frozen spectral layers from your input, stack them into a sustained cloud, and Morph / Smear / Jitter through it. Stacking many layers is the job that genuinely suits a GPU, so there's an opt-in GPU engine for it.

Honest about where the GPU helps

A few layers are faster on the CPU (the default). The GPU engine wins at big clouds — at 128 layers the CPU overruns the real-time budget (11,520 µs/hop) while the GPU does it in 1,946 µs (5.9×). The live status line shows the engine, layer count, measured µs/block, and a % of real-time headroom gauge derived from your own device.

Verified

GPU output matches a CPU reference bit-for-bit (xcorr = 1.0); the GPU's high-layer timing is impossible on the CPU; a sustained GPU load raises the machine's GPU-utilization counter well above idle.

One notarized installer (Customize pane → AU / VST3 / CLAP / Standalone). Apple Silicon; run your host natively (not Rosetta).