Skip to content

v1.2.8

Choose a tag to compare

@github-actions github-actions released this 15 Sep 21:03

Download

Ready to run - the filters are already inside. Unzip, run aura-engine.exe, drop a track on it. The app opens on the filter that came with it.

Bundle Filter Output rates Size
Compact 5k taps every FS multiplier (FS2-FS16) 7 MB Download
Starter 1M taps every FS multiplier (FS2-FS16) 134 MB Download
Standard 10M taps FS8 (352.8 / 384 kHz) 326 MB Download
Reference 30M taps FS8 (352.8 / 384 kHz) 966 MB Download

Unzip more than one into the same folder and the app offers every tap count you have.

Already have the filters, or want to build from source? The plain app zip is below, and the individual filter packs are on the 1.0.0 release.

Checksums (SHA-256)

7b7589ce3246e70de82f60703aaaedd785610e245f276e35544e79371e5486fe  aura-engine-v1.2.8-bundle-5k-windows-x64.zip
1d32666bdf2a9013b6767c3f1bee773c88f98a2e3db5985293b0f8ae78f227d1  aura-engine-v1.2.8-bundle-1M-windows-x64.zip
092a327307d624c61d992016742668a4d9aa5f07132409bc80ae025ac1ecc06f  aura-engine-v1.2.8-bundle-10M-windows-x64.zip
20786a9fe671c491769ee92bb546635cfbebf59dd997ba05b36c9201d6ba7d86  aura-engine-v1.2.8-bundle-30M-windows-x64.zip

What changed in 1.2.8

Subsonic filter.

In the Audio Science Review thread a listener showed a CD whose spectrum rises
into a hump below 20 Hz. It is not an offset: at 65 536 bins it sits at about
1 Hz, at close to 1 % of full scale, and differs between the channels. Below
20 Hz the converter did one thing — it subtracted each channel's mean — and
that takes 0.00 dB out of a wave like this one. 1.2.8 adds a switch for it.

It is there for speakers, not for sound. Nothing here claims it is audible.

What it does

A linear-phase FIR high-pass at the source rate, straight after DC blocking,
before the apodizer's forensics and before the filter. Off by default.

corner 20, 15 or 10 Hz — the chip on the SUB row
passband flat within 0.01 dB from the corner up (designed: 5·10⁻⁵ dB)
stopband at least 100 dB down from half the corner to DC (designed: 104.4 dB)
−6 dB at three quarters of the corner
length from the source rate: 29 813 taps at 44.1 kHz and 20 Hz, twice that at 88.2 kHz or at 10 Hz
filename SUB20 / SUB15 / SUB10, before AA and HP

Three corners, each for a reason: 20 Hz is the edge of the audible band and
the classic subsonic corner; 15 Hz keeps the lowest note of a 32-foot organ
stop, 16.35 Hz, flat; 10 Hz takes drift and infrasound and nothing musical.
There is no corner above 20 Hz — that would be cutting bass.

Linear phase is the reason to do this offline. Above the corner nothing moves
in level or in time, which a high-pass in the playback chain cannot promise.

What was measured

Through the whole app, 5k taps at FS8, each file converted with the filter off
and on, with the dither seeded identically in a test build so the two outputs
can be subtracted.

A synthetic file built like the one in the thread — 1 Hz at −40.5 / −45.0 dBFS
under music-like noise from 30 Hz up and a 40 Hz tone: with SUB20 the 1 Hz
component comes out at −158.8 / −163.3 dBFS. Above 25 Hz the difference between
the two outputs stays at least 100.8 dB below the signal in every 0.34 Hz bin,
the cross-correlation lag is zero, and the 40 Hz tone has the same amplitude
and phase to five decimals.

A CD rip (Santana, Mal Bicho): above 25 Hz the difference stays at least
98.9 dB below the signal; below 8 Hz the level falls from −51.5 to −142.6 dBFS.

With the switch off, the output is byte-identical to 1.2.7 — SHA-256 equal
on that track and three synthetic files, on CPU and GPU, with the Adaptive
Apodizer and Hybrid-Phase on and off.

What to know

  • The output ceiling is measured on the finished render, after the filter.
    Taking out a large infrasonic component can move the peak, and the ceiling's
    scalar with it: −0.027 dB on Mal Bicho.
  • With Hybrid-Phase on, the output above 20 Hz is not identical. Hybrid-Phase
    switches between its linear- and minimum-phase renders at zero crossings, and
    removing a slow wave moves zero crossings, so some switch points move. Where
    one does, that stretch comes from the other branch. The difference is
    Hybrid-Phase's own and it is local: on Mal Bicho 89 % of 10 ms windows still
    differ by less than −80 dB relative to the signal.

Advanced DSP, as a rack

The three checkboxes and the new stage are lit rows now, top to bottom in the
order the pipeline runs them: SUB, AA, PFR, HP. A row lights in its stage's
colour when it is on. Your saved settings carry over.

The line under the heading used to append every option that was on and wrap
wherever it happened to, so the panel jumped each time one was switched. It is
two fixed lines now: the filter on top, the output below, with the tap count
written as the filename writes it — 30M, not 30.0M.