What's new

Why we rebuilt it

An audit of 0.4.5 found four defects in the signal path. Rather than patch them one by one, we replaced the engine and wrote a test for each defect so it cannot come back.

• →The limiter changed its gain in steps instead of smoothly, which added distortion on every transient.
• →The glue compressor ignored its attack setting and clamped down instantly.
• →The loudness stage was a gain boost into a soft clipper, aimed at −7 LUFS by default.
• →Saturation ran without oversampling and aliased on bright material.

The new engine

Ten modules in a fixed order: gain, restoration, match EQ, parametric EQ, dynamic EQ, compressor, multiband, saturation, stereo, and the limiter. Every module can be bypassed or edited, and every one can be reset to the proposed value.

• →True-peak limiter with lookahead, dual release and 4× oversampled detection. The ceiling is yours to set; −1 dBTP by default.
• →Oversampled saturation that adds warmth without aliasing.
• →A compressor whose attack and release do what the labels say, plus multiband compression and dynamic EQ for problems that only appear in some bands or some moments.
• →Match EQ toward a target curve, or toward the curve of a reference track you drop in.
• →Restoration (de-noise, de-click, de-hum, de-ess) is proposed only when the analysis is confident there is something to fix, and stays off otherwise.
• →Width and balance changes are suggested but left off until you enable them. The only automatic stereo action is folding the low end to mono when it is badly out of phase.

Findings, not scores

There are no quality scores or tiers anywhere in the app.

• →The analysis lists what it measured: tonal balance, dynamics, loudness, true peak, stereo image, noise floor, clicks, hum, sibilance and clipping, with the numbers.
• →Each finding says which modules it proposes, and each proposed module carries the reason in plain words, so you always know why a step is in the chain.
• →Chain cards show their main controls inline and, after a preview, what each module actually did, such as the gain reduction on the dynamics modules.

Hear it before you commit

• →A 20-second preview of the loudest part of the track renders through the current chain in about a second and plays A/B against the source.
• →Drag the loop region anywhere; the preview re-renders when you change a control.
• →Optional codec preview: hear the segment through MP3 or AAC encoding and see the true peak after the codec, which is commonly half a decibel to a decibel above the PCM ceiling.

Loudness and delivery

• →Presets: Dynamic (−14 LUFS), Balanced (−11, the default), Dense (−9), Club (−7), Broadcast (−23), Podcast (−16), Spotify (−14), Spotify loud (−11) and Apple Music (−16).
• →A platform table shows the gain each streaming service applies to your master on playback, and a readout estimates how much limiting the chosen target needs before you render.
• →Export to WAV, FLAC, AIFF, MP3 or M4A, with sample rate, bit depth and metadata (title, artist, album, year, ISRC) in one dialog.
• →Proper dither (TPDF or shaped) on 16-bit output, and the true peak is re-checked after sample-rate conversion so the delivered file respects the ceiling.
• →A mastering report with before/after measurements (integrated loudness, loudness range, peak-to-loudness ratio, true peak, stereo correlation), a per-module summary, and a delivery check of the exported file. Export it as a printable HTML page.

Faster, and better behaved

• →Analysis of a full-length track that took about a minute in 0.4.5 now takes a few seconds.
• →The engine compiles its processing kernels once and caches them, so previews are quick from the first one.
• →A force-quit no longer leaves the engine running in the background and blocking the next launch.
• →The local engine accepts requests only from the Tunr app itself.

Intel Macs

• →Tunr now ships two macOS installers: Tunr.dmg for Apple Silicon and Tunr-Intel.dmg for Intel. Both are signed and notarized by Apple and run on macOS 11 or later.
• →They are separate builds rather than one universal app because the numeric libraries behind the engine are not available as universal binaries.
• →The download page shows both; the in-app updater picks the right one for your machine.

Windows

• →The Windows installer, Tunr-setup.exe for 64-bit Windows 10 or later, is 0.5.0 as well.

Honest limits

• →Very loud targets such as Dense and Club can stop short of the target when reaching it would hold more than 30 % of the samples in gain reduction. The report says so, and by how much. Enabling a compressor or multiband stage before the limiter usually closes the gap.
• →True peak is measured with a 4× oversampled meter as in ITU-R BS.1770. Ceilings above −0.5 dBTP can read a little higher on 8× or 16× meters.

Updating

• →Installed copies of 0.4.x update themselves through the in-app updater. Licence keys are unchanged and keep working.