v1.3.2

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)	135 MB	Download
Standard	10M taps	FS8 (352.8 / 384 kHz)	327 MB	Download
Reference	30M taps	FS8 (352.8 / 384 kHz)	967 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)

fe86b727274e9e1453a10fe1eab479448d3fc97dcd9d61ec1b8792e19d0f4b07  aura-engine-v1.3.2-bundle-5k-windows-x64.zip
08952d6295d49be5a2bddeed3679fc00aa31a8f545dfdee596ebf369cfff50d5  aura-engine-v1.3.2-bundle-1M-windows-x64.zip
1c50af9205202795928b60f4e6207da664a33c93469385a3fe28a3562589f458  aura-engine-v1.3.2-bundle-10M-windows-x64.zip
cf157df65e3f75983e741416760217b4ff20ba291aee7738bed9d827fad6c692  aura-engine-v1.3.2-bundle-30M-windows-x64.zip

What changed in 1.3.2

Hybrid-Phase stops clicking. Only files converted with HP on are affected.

Erogen, in the Audio Science Review thread, posted a spectrogram of a 1.3.1
file and asked what the thin vertical lines above the apodizer's 18 kHz
ceiling were. They were Hybrid-Phase: every switch between its two branches
left a tick. Thanks for spotting it.

Why it ticked

Hybrid-Phase renders the track twice, in linear and in minimum phase, and
changes branch around each attack. The switch sample is snapped to a zero of
the mid difference between the branches, so both channels change phase at
the same instant. That leaves a small step in each channel, and the 0.083 ms
fade meant to hide it was far too short to do anything but turn it into a
broadband tick.

On a clipped CD at 176.4 kHz / 30M taps, content above 24 kHz peaked at
−42.9 dBFS. Subtracting a pure linear-phase render recovers exactly the 1070
switch points the log reports, and all 895 ticks above −80 dBFS peak within
0.1 ms of one of them.

The fix

The fade is now 1 ms, on the running integral of a Hann window instead of a
raised cosine. The switching logic is untouched: same switch points, same
snap, same 20 ms hold.

	1.3.1	1.3.2
above 24 kHz, 176.4 kHz / 30M	−42.9 dBFS	−103.0 dBFS
above 24 kHz, 352.8 kHz / 1M	−43.8 dBFS	−100.7 dBFS
switch points	1070	1070
true-peak gain	−3.74 dB	−3.74 dB

What it costs

For that millisecond both branches sound together, and at the very top their
phases disagree. Against a click-free splice, 14–21 kHz in the 3 ms around a
switch comes out 0.32 dB lower (median), 8–14 kHz 0.05 dB lower, and nothing
changes below 8 kHz.

Who it affects

A fresh install has opened on TFS Phase, with HP off, since 1.3.0. HP stays on
for anyone who carried their settings over from 1.2.x. Conversions without HP,
with αHP or with TFS are the same arithmetic as in 1.3.1.