Skip to content

v0.5.2

Choose a tag to compare

@github-actions github-actions released this 12 Oct 05:22
· 226 commits to main since this release

Download

Installer
Windows FSVR-Windows-Installer.exe Run it; it asks for admin rights to install into Program Files
macOS FSVR-MacOS-Installer.zip Unzip and open FSVR-MacOS-Installer; it is not signed, so right-click it and choose Open the first time
Linux FSVR-Linux-Installer chmod +x FSVR-Linux-Installer, then run it

The installer puts every format in place (VST3, CLAP, VST2, the 32-bit VST2 with its DXi, AU and the standalone) and can keep FSVR up to date. To install one format by hand instead, take its own zip from the list below, FSVR-<OS>-<format>.zip. FSVR-MacOS-Update is for the updater, not for downloading.


FSVR v0.5.2: the editor redone, and effect parameters that stop ringing.

This round answers jameshansen's redesign of the editor, mkruselj's reports on the keyboard and shortcuts (#6, #11), dezzo's on Cubase and host windows, WouterVanNifterick's operator transpose (#31), and a report from Reaper that the variation effect fed back for good once FB Level was touched. mkruselj, dezzo and WouterVanNifterick also contributed the fixes for their own reports.

Reverb and variation parameters write both bytes. Reverb P1-P8 (10 00 50 to 5E) and variation P1-P12 (10 00 68 to 7E) are two-byte words, but the engine's test for them returned before reaching those addresses, so every change to one wrote its low byte into the word's high byte. The word then clamped to the top of its range: a chorus or flanger's FB Level became +1.0 and rang with no notes playing, whatever it was set back to, and a phaser or flanger picked afterwards read the same bad words. This has been in every release since v0.4.1. A project saved with one of those parameters touched keeps the bad byte; touching the parameter again under v0.5.2 rewrites both.

The editor redone. Every image is rendered in Blender at each of the window's scales, 0.5x to 2x, and every caption is drawn at run time from the skin's fonts. An Operators tab holds the seven expert pages as sub-tabs; the Parts page is rebuilt with steppers on every number and radio pairs for every two-way choice; the curves sit on dark screens with antialiased lines and round handles. The operator panel shows the operator's own waveform and its first 16 harmonics, computed by the engine. An Options tab gathers the scales, the two keyboard layouts, Init Patch, Turn off all Effects, Init All, Quality / Speed and the standalone's audio and MIDI. The envelopes' boxes are tagged T1 to T4 and L1 to L4 as the unit names them, with stage names as an option (Modern EG Labels). The expert pages' icons are Fontaudio's, by fefanto (CC BY 4.0), and each group's title tab now stands on its box's square left corner.

The EG plots draw the engine's own curves, with zoom, pan and a synced overview (WouterVanNifterick).

A formant operator's Transpose moves the operator's pitch, the grain rate, and leaves the formant's centre where Coarse and Fine put it; the operator column shows Transpose beside a Fixed mark when the form is Formant (WouterVanNifterick, #31). He checked it against his own unit. It has not yet been retested against a capture here, and STATUS.md lists that retest as open work.

Hosts. An edit made in the editor stays in the VST3 controller, so Cubase stops undoing it, and the editor follows the host into a new window instead of staying in the old one (dezzo). Keyboard shortcuts open the editor's pages, also in hosts that take keys off the window, and text entry gets its characters (mkruselj, #11). The on-screen keyboard shows notes on and off, never what the sustain pedal holds (mkruselj, #6). POLY counts sounding channels rather than keys held (mkruselj). The Windows standalone builds the ASIO back end, and the Audio and MIDI dialog has ten MIDI input rows instead of four (mkruselj).

A part's voice name comes back after a session restores it, and the LCD's channel ends where CPU's does.

Research. The VOP3 effect chip's instruction set was measured on rgwan's unit over sessions 8 to 39, and a C++ core runs the measured rules (src/fs1r/chips/vop3_core.h, vop3_modules.h): Hall1 and Hall9 click replays match the unit to rounding. It is not in the render path yet; the effects are still the model.

Checks. build.bat test passes (effects, engine self check with a new check that reverb and variation words take both bytes, check_formant, check_presets), and build.bat plugin passes check_plugin and check_gui. Against a fingerprint of v0.5.1's own build, tools/regress.py renders 18 of its 19 cases unchanged; ctrl-multi (performance 28) moved with the transpose change, centroid 171.2 to 167.5 Hz and peak count 4 to 5, and nothing else moved it. The tracked regress_ref.json is still the stale one. The macOS and Linux builds were not run locally; CI builds them.