v1.5.0

What's new in 1.5.0

The converter plays, and tunes in. Every file in the list has ▶ next to C, and the player runs the chain the file would be converted with — so a setting can be heard before a batch is spent on it. Internet radio plays through the same rack. The full story, with measurements, is in the CHANGELOG.

Added

• A player on the conversion's own chain: ▶ on any file plays it through the rack it would be converted with; C converts it
• Internet radio through the same rack: featured stations, favorites, the Radio Browser catalog sorted by name, votes, plays or quality and narrowed to lossless or a bitrate, or any stream address; MP3, AAC, HE-AAC, FLAC, Vorbis, Opus and HLS; BIT-PERFECT on a stream too
• Album Level (on by default): the queued tracks of one album — same folder, same Album tag — share one true-peak gain, so the album keeps its balance
• Analyzer: loudness, LRA, true peak, DR and RMS for the source and the output, and a spectrogram that compares them — over a whole file, or live on a radio stream, song by song
• Visualizations: GLSL scenes, a studio for writing your own, and scenes that follow each instrument (a 282 MB model pack, downloaded on first use), on a file and, with a graphics card, on a radio stream
• Updates install themselves from this release on, after their signature is checked
• Instant start in the player's menu — switch it off, and the first sound already has every stage in it
• The graphics card takes over playback when the CPU cannot keep up in real time
• A seek crossfades into the new place, and the next track follows without a gap
• Badges under the player show which stages are already in the sound; a thin bar fills while they switch in
• BIT-PERFECT, per-track settings (M), a playlist with repeat, drag to reorder, and output device and volume
• Each converted file shows how many times real time it took; the batch's figure is on Convert all and in the session log
• Listening mode (L): the settings fold away; the album cover fills the player when nothing is touched
• A short tour of the window on first start, and again under the ? below the version number
• THIRD-PARTY-NOTICES.txt: the libraries the program is built with and their licences

Changed

• Dropping files adds them to the list instead of converting them at once: C converts one, Convert all converts the rest
• One copy runs at a time: the player needs the output device to itself, so a second launch brings the first window forward. Running two copies side by side to convert faster no longer works; a single copy that keeps up with two is in the works
• TFS is derived as designed: its files differ from 1.3.x by at most −120 dBFS in the audio band, and its stopband is −193 to −231 dB instead of −94 to −126
• Declip leaves MP3, AAC, Vorbis and Opus sources as decoded: what it took for clipping there was the codec's own overshoot
• The Adaptive Apodizer stands down where its cutoff would sit at or above a codec's wall, and says why
• The settings sit in one strip of five cells; the list shows ten rows
• The Window list is gone: the filter has always been Kaiser, and the list only changed the file name
• The convolution keeps half of each spectrum, on the CPU and the graphics card: long filters need half the memory, and a 30M ×8 conversion on an RTX 4090 runs up to 1.4× faster
• The Subsonic Filter runs on short partitions: less memory and time, the same result to −297 dBFS
• Every window has lost its system frame and reopens where it was closed; the audit-log console starts minimized
• A press outside an open menu, panel or drop-down list only closes it and no longer reaches what lies under the pointer
• The thread pool that conversions and the player's preparation share runs below normal priority, so a batch does not starve live playback
• A new app icon; building from source needs Rust 1.88 and CMake

Fixed

• Sources at 88.2 kHz and above were cut at the CD's wall (about 21–23 kHz); each source rate now gets the filter for its own ratio
• Converting at FS16 (705.6 / 768 kHz) stopped when the FLAC was written; it is written now (a player with an older FLAC decoder may refuse a rate that high)
• Writing FLAC could take the whole program down on some signals (full-scale 24-bit noise): such a frame is now written uncompressed
• HE-AAC files were decoded without their top octave, or not at all from .m4a; they are now decoded at full rate inside the program, on every edition of Windows
• Chained Ogg files stopped at the end of their first stream
• Two long files no longer end up on one 8 GB card together: a long file reserves the card for its whole filter set, and the next one waits
• A long file checks for room for its temporary files before it starts, instead of failing minutes in
• The Headroom tooltip described the control as it worked before 1.2.3, and Off read "0 dB"; Off is now marked "−0.5 dBTP, as shipped"

---

Upgrading

From 1.3.4 or earlier: those versions have no updater. Download 1.5.0 below and unzip it over your old folder, or anywhere else — your settings are kept. From 1.5.0 on, new releases arrive inside the app.

Files converted with TFS (the default) come out slightly different. The filter is now built as designed; the first TFS conversion at each tap count derives it again once (at 30M taps: about 12 s and 4.5 GB of free RAM).

Album Level is on after the update. Albums whose tracks needed different true-peak cuts now come out at one level, which is lower on those albums than before. Switch ALB off in the rack's Output zone to convert track by track as in 1.3.4.

Dropped files wait for you. A drop fills the list; C or Convert all starts the work.

---

Download

Ready to run — the filters are already inside. Unzip, run aura-engine.exe, drop a track on it: ▶ plays it, C converts it.

Bundle	Filter	Output rates	Size
Compact	5k taps	every FS multiplier (FS2-FS16)	12 MB	Download
Starter	1M taps	every FS multiplier (FS2-FS16)	140 MB	Download
Standard	10M taps	FS8 (352.8 / 384 kHz)	332 MB	Download
Reference	30M taps	FS8 (352.8 / 384 kHz)	972 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. The visualizations with instruments fetch their model pack by themselves.

Checksums (SHA-256)

90ae91f1c8f1cdd293871a4bf8a6814ddc6f76df69baa78f8439640e5a7ebb98  aura-engine-v1.5.0-bundle-5k-windows-x64.zip
25ecd897a7e121ea1d86dc93461a76295b1e4125d1e5b49c268f454fa56715ae  aura-engine-v1.5.0-bundle-1M-windows-x64.zip
3637bfb6ccc7418c6bc0cd2ff3d2fbc72843eed73fbc162a6f540be954a54b95  aura-engine-v1.5.0-bundle-10M-windows-x64.zip
85c1b518b4c98e792f3e16158b4d84b20dbf69f2baf7ea276db0c07d62d641a4  aura-engine-v1.5.0-bundle-30M-windows-x64.zip
22c620196eae92ac7b67c7725f29246c09d444e45b9705792b2e0fd7a65adff7  aura-engine-v1.5.0-windows-x64.zip

Requirements

• Windows 10 or 11, 64-bit; a CPU with AVX2; the WebView2 runtime (Windows 11 has it; on Windows 10 the app offers the download if it is missing)
• For playback: the output device is opened in WASAPI exclusive mode at the output rate, so nothing else can play through it at the same time, and it must accept that rate (FS×8 is 352.8 / 384 kHz; otherwise choose a lower FS multiplier)
• For radio: an internet connection
• A graphics card is optional: Vulkan with SPIR-V passthrough for conversion; playback uses a discrete card, and only when the CPU cannot keep up; the scenes with instruments want 3 GB of its memory free, and on a radio stream cannot do without it

Thanks to masuplc2, whose tables in Discussions #5 made the case for Album Level, and to Erogen on Audio Science Review for the report on two long files and the question about batch speed. The station catalog is Radio Browser, an open database made for players like this one.