Release notes and limits
Windows x64 release collection · 31 October 2026. Public beta is complete following owner testing.
ANUBIS coordinated Windows plugin collection — 31 October 2026
All eleven native automated test groups passed: processor behavior across sample rates/block sizes; SERQET reconstruction and repair; SOBEK alias case; offline timing measurement; saved noise profiles and music-relative riding; full-length partitioned IR against direct convolution; generated SIA calibration; all 48 DUAT patches; TEFNUT alignment improvement/stereo rendering; ANPU and MERET DSP.
All fourteen newly packaged VST3s loaded and rendered finite audio through a native offline VST3 host. State save/restore was exercised. NEITH also accepted and rendered its external input bus. Existing eight public plugin binaries were used for their demonstrations.
SIA integrated loudness/LRA were cross-checked with FFmpeg 7.1 on generated material. A subset of locally synthesized EBU Tech 3341 calibration definitions (steady loudness cases 1–2 and true-peak cases 15–19) passed their specified tolerances. This is not the full official corpus or conformance certification. Broadband-noise/near-Nyquist differences against FFmpeg remain documented in metering-comparison.json.
Automated and offline checks do not establish real-vocal quality, physical interface performance, compatibility with every DAW, or long-session reliability. These are unsigned releases. Begin with a copy of the session. No macOS, AU or AAX binaries are included.
DUAT
48 factory patches and 24 voices. Original electronic synthesis, not acoustic recordings. Factory ADSR settings are compiled in. Tone/character/movement/output edits recall with the DAW project. No external sample or preset-bank import. Synth drums use their selected patch character, not a General MIDI map.
ANPU
Zero added buffering latency. Detector filters introduce phase shift. Start in Processed monitor mode and adjust threshold for the recording.
MERET
Zero added buffering latency. Split mode uses a first-order residual, not a linear-phase crossover. No automatic makeup gain or limiter. Return Monitor to Processed before export.
SERQET
1024 samples host-reported latency. For a learned profile, play noise only with Capture noise profile enabled, then disable capture before processing the wanted signal. Profiles recall at the same sample rate; recapture after changing project rate. Short isolated clicks and plateaus only: this is not reconstruction of long missing or badly clipped audio.
SOBEK
16 samples host-reported latency. Filtered 2x processing reduces aliasing; it does not eliminate it. Optional RMS matching changes gain gradually.
SET
Optional sample-peak guard clamps samples at 0.99; it is not a true-peak limiter. RMS matching can reduce the apparent gain difference of shaping. Defaults preserve a neutral signal.
NEITH
Complementary first-order residual crossovers are minimum phase. Use the host sidechain input and select External sidechain for external detection. Missing external input falls back to the main signal. Expansion is downward expansion with bounded attenuation.
SHU
Measurement compares the first second, up to +/-100 ms, and is intended for related recordings of the same event. A correlation below 0.6 is rejected. An early target can be delayed automatically; a late target requires delaying the guide or moving the target earlier in the DAW. This is not automatic alignment of unrelated performances.
SIA
Mono/stereo only. Measures input before reference audition. Reference must match project sample rate for playback and must be 0.4 seconds to 5 minutes. Reference audio is not embedded in project state: reload the local file after reopening. Matching is bounded +/-12 dB. LRA is provisional on short segments; use a full program. True peak is an estimate. Generated calibration cases pass, but broadband and near-Nyquist comparison differences remain; this is not an EBU/ITU-certified delivery meter. Turn reference listening off before export.
RENENUTET
An RMS rider with high-frequency boost restraint, not a trained voice/breath classifier. Music-relative mode requires an enabled host sidechain; without a usable music level it uses the fixed RMS target. Gain range and riding speed remain bounded.
NUT
Host BPM drives tempo sync. If the host supplies no BPM, the default is 120 BPM. Flanging and phasing can intentionally affect mono compatibility. Set Wet mix to 100% for pure vibrato.
TEFNUT
Live ADT adds an intentional wet delay; no compensation latency is declared. Offline guide/dub export accepts related mono/stereo files at the same sample rate, up to 60 seconds. Stereo channels share the same time map. Grain-based timing edits can introduce artifacts; audition the exported WAV. It does not correct pitch, tune harmonies, or align unrelated material. Export to a new file.
TAWERET
Separate product and VST3 identifier from BES Voice Workstation. Includes an original synthesized kick. Load a mono/stereo drum sample up to 10 seconds; stereo imports are averaged to mono and embedded in the session. Triggering uses an envelope threshold and retrigger hold, not a trained drum classifier.
MONTU
272 samples host-reported latency (about 5.67 ms at 48 kHz). Original nonlinear amp characters, not measured circuit emulations. Loaded IRs are averaged to mono, resampled at import and limited to 16384 project-rate samples. Reload the IR if project sample rate changes. Partitioned convolution and filtered 2x amp processing are included. No recorded commercial cabinet library is bundled.
BES Voice Workstation
The coordinated build uses standard Windows audio backends in standalone. In a DAW, the host handles your audio-interface driver. The separately downloaded recorded Foundation Bank is required. Live singers, microphones, physical interfaces and broad host compatibility still need testing. See its included bank licences and guide.