How the options work
The options come in two groups. MIDI processing works on the notes, timing and controllers of the MIDI/KAR; Real-time sound and mix works on the sound already produced by FluidSynth. In both cases the original file is never modified and no audio files are created: the improved MIDI is a temporary copy used only while listening.
- Each checkbox is tied exactly to its function: ticked = applied, unticked = ignored.
- The options apply to MIDI and KAR. They have no effect on video, MP3 and CD+G.
- Pressing Apply during playback, the sound effects change immediately; the MIDI options apply from the next Play.
- The options are saved per song; Save as defaults uses them for all new songs.
MIDI processing
The main window adapts its height to the monitor work area when opened. Options uses the wide layout: MIDI processing on the left, sound and mix on the right, buttons at the bottom. Layout is set when opening the window and does not automatically change when moving it to another monitor.
Clean up anomalous events and prevent stuck notes
What it is for: fixes the most common defects of downloaded MIDI/KAR files: notes left «hanging» (note-on without note-off, sustain not released), missing tempo, time signature and key signature, and average velocities that are too high or too low (if the average is above 110 or below 25 the velocities are brought back into that range). At the end of the song an all-notes-off message is sent.
Risk: very low. It changes neither the notes nor the lyrics.
Specific musical corrections per instrument family
What it is for: each family receives only the correction suited to how it is played. For winds and accordion/harmonica the pitch bends are halved (less «ribbon-like» slides), very short expression gaps (CC11) are removed and the expression volume is kept between 50 and 127, so the sound neither disappears nor deflates. For basses, guitars, strings, voices and pads the corresponding rules apply (for example the bass pitch bend always stays centered; on guitars the tiniest pitch wobbles, up to 25 cents, such as certain vibratos written in the file that make the guitar sound flat, are brought back to center, while real bends up to about 3 semitones are kept; long «slides» written with the pitch wheel that drop the note by more than 3.5 semitones are brought back to center) and for all non-drum instruments overlapping repeated notes are fixed.
Peak limiter and context (01/11): on strings, pads and winds, notes standing out far above the part’s own median level are lowered by at most 6 dB (CC11 and velocity); GM pads 88–95 stay 4.5 dB lower as background. With Balance also active, each channel is compared with the others through a table of loudness per GM instrument measured for the default SoundFont (airfont_380_final.sf2): a part that is too loud is lowered (max 6 dB), a part that is too quiet may be gently raised (max 4 dB). With other SoundFonts the table is an approximation. The lead channel, drums and muted channels are excluded; nothing is written to the file.
Guitars (16/11): the sustain pedal (CC64), a keyboard gesture a guitar does not have, is removed: notes end at their note-off and the natural tail comes from the sample. In acoustic guitar arpeggios (nylon and steel) the string keeps ringing until the same note is played again, so notes do not sound dry. Overdrive and distorted guitars (GM 29–31) are levelled: inside a chord no note rises above the average of the others, and the first chord after a rest longer than a bar (or at the start) no longer comes in louder than the following ones. The Guitar correction strength slider (1–10, default 5) sets the levelling: 1 almost off, 10 full. It only works downwards, never raises levels, and acts while this option is on.
- It applies after the style: if the style changes the instrument of a channel, the correction follows the new instrument.
- It coexists well with Humanization: winds are not lengthened by humanization precisely so as not to spoil the expression corrected here.
- Turn it off if the MIDI has hand-crafted expression you want to hear untouched.
Slightly humanize velocity and performance
What it is for: makes the performance less mechanical. Velocities are pulled toward half of their value and randomly varied by a few points (±3; drums less), then notes (drums excluded) are lengthened up to the next one, within a short limit, for a more legato sound. Winds are not lengthened. The result is reproducible: the same song always gives the same result.
- Electronic style (or dance/techno): the regularity of the machine is intended, turn humanization off.
- With Groove both act on the «feel», but on different channels: groove on mechanical channels, humanization on all. Together they work well; if the song sounds «soft», turn one off.
- Together with Piano dynamics it works in the same step: no problem.
- With Balance and Light compression the dynamics are leveled three times: if the song loses breath, turn the compression off.
Finger slides on strings (01/11): with Humanization and the guitar corrections active, the player replaces the GM 120 noises written in the MIDI with its own, placed only on real finger slides of acoustic/clean guitars (explicit portamento, melodic shift leaving the recent notes, chord shape moving along the neck). They are inserted into the MIDI on a free channel with a level relative to the guitar and a short reverb, so they sit in the mix. The bank must contain the GM 120 preset. The vocal and the melody are never touched.
Humanization and pedal: added note-release extensions cannot cross sustain, sostenuto, reset or note-stop events. This prevents an already released note from being captured by a later pedal press. Original pitches, onsets, controllers and authored overlaps are preserved; controllers on other tracks of the same channel are included.
↑ indexApply light groove to mechanical parts
What it is for: detects the «mechanical» channels (at least 16 notes and at least 85% at the same velocity, typical of sequencers and backing tracks) and adds a light groove to them, so they do not sound like a machine.
- Electronic and dance songs on a rigid grid: turn it off.
- Slow songs or songs already played by a musician: not needed, and it finds nothing to correct.
- The style quantizes first; the groove then puts a little life back into the quantized notes.
Balance levels between channels and instrument families
What it is for: analyzes the MIDI and recalibrates the velocities of channels that are too loud or too quiet relative to the others, so that no instrument covers the others.
It works on the note velocities (not on the volume rendered in LUFS as in CreaVideoLyrics, because here no audio is created).
Context in the final mix (01/11): with Balance active, section strings (GM 48–54) and the muted guitar (GM 28) are compared with all the other instruments after their corrections and lowered if they stand out (12 dB under the sum of the others, max cut 12 dB). If the drums/bass reach their maximum cut and are still too loud, the leftover (max 6 dB) is taken from the loudest guitars. With a convolution reverb the internal reverb is switched off automatically, so there is no double reverb.
Context between channels (01/11): in addition to velocities, strings, pads and winds are compared with the other channels through the per-instrument loudness table of the default SoundFont (airfont_380_final.sf2): too prominent parts are lowered via CC7 (max 6 dB), parts that are too quiet are raised cautiously (max 4 dB). The estimate is most accurate with the default bank.
Measuring ahead of time (07/11): the balance measurement is made once per song, in the background at low priority, as soon as you select the song or open/extend a playlist (up to 500 songs). It starts from the original MIDI, so it does not change with Retouch and the other options; the result is kept between sessions. This way playback starts immediately when you press Play. With rearrangement active, the processed song is measured when Play is pressed.
- With duplicated parts it works in the same step: echo copies are first attenuated, then balanced.
- With Leveling and Light compression the volume is handled at several points: if the mix sounds flat, keep only balance and leveling.
- The mute removes a channel before the analysis: the balance does not count it.
Detect and attenuate parts duplicated at the unison or octave
What it is for: many MIDIs repeat the same melody (for example a sax in echo) on several channels, and the sound doubles or goes out of tune. Here the copies are recognized and attenuated; pan and instrument of the echo copies can be replaced with those suggested by the analysis.
- If the octave doubling is intended (strings and brass supporting the melody), the arrangement becomes sparser: turn the option off.
- With channel Balance they are meant to be used together, in the same step.
Turn choirs into strings
What it is for: the synthetic choirs in many MIDIs (Choir Aahs, Voice Oohs, Synth Voice) often sound fake. With this option they become strings (String Ensemble 1). The original MIDI file is not modified. It does not apply with an external MIDI output, which has its own choirs.
- With ensemble doubling the resulting strings are doubled like all the others.
- If the choirs are an intended, characteristic part of a song, turn the option off.
Double strings, pads, organs, accordion, brass, sax, choirs and guitars
What it is for: while the song plays, a second synthesizer adds a «shadow» voice to some instruments. There are two modes, depending on the instrument. The MIDI file is not modified: the shadow exists only while listening.
1. Ensemble doubling (strings, synth strings, choirs and voices, organs, accordion, violin, pads): the same part is also played by a second voice, slightly detuned (4 to 11 cents), quieter and on the opposite side of the stereo space. The original channel drops by about 1.5 dB to compensate for the sum. This is the ensemble effect.
2. Timbral reinforcement (guitars, brass, sax, reeds and flutes): useful when the SoundFont gives these instruments little presence. Under the original instrument plays, quieter (40-50%) and at the same point of the stereo space, a complementary instrument that gives it body, with a difference of only 2 cents. The volume of the original does not change. Examples: clean electric guitar + steel guitar, nylon guitar + steel, jazz guitar + nylon, overdrive + clean, alto sax + tenor, tenor sax + alto, trumpet or trombone + horn, brass section + synth brass, oboe + clarinet, flute + pan flute.
Never doubled or reinforced: drums and percussion, basses, pianos, harp, vibraphone, muted guitar and plucked instruments.
Notice: once per song, after about half a second of playback of a MIDI/KAR, a window appears listing the channels with reinforcement or doubling (for example «Channel 3: clean electric guitar + steel guitar»). The music continues while it is open.
- In ensemble doublings, if two channels use instruments of the same family (for example two string tracks) the shadow is added to only one, so as not to sum parts that already overlap. In reinforcements this rule does not exist: each guitar or wind gets its own, up to a maximum of 6 reinforced channels together.
- The reinforcement neither turns off nor replaces the instrument of the file: if you do not like the result on a song, untick this option.
- With an external MIDI device chosen (see External MIDI output) doubling and reinforcement are disabled: only the expander plays.
- The option has a single switch for both modes: unticking it turns off ensemble doubling, reinforcement and the notice.
- Doubling uses more CPU (the second synthesizer plays in parallel): on slow computers, if the sound «skips», turn the option off.
- It applies at the next Play; while listening, a change of key, style or mute also restarts the song from the current point with the new setting.
- With Chorus and Stereo widening the width adds up: if the sound becomes too wide and confused, turn one of the three off.
- With duplicated parts it works in a complementary way: that one attenuates copies already present in the file, this one adds one where the part is single.
- If the song already has many ensemble instruments the sound becomes dense: turn it off.
- With the bass, low-mid and treble boosts, guitars and winds can become too full or harsh: judge by ear.
Breath of the winds
What it is for: winds are the weak point of MIDI: every note starts and ends the same, and the sax sounds like an organ. Here sax, single brass (trumpet, trombone, tuba, horn), oboe, bassoon, clarinet and flutes receive a human touch, on single notes only (never on chords or sections):
- Breath: the note starts a bit quieter and grows, the more so the longer it is (expression CC11).
- Natural vibrato: only on long notes, delayed, growing slowly, a few cents wide, with frequency and amplitude that change from note to note and within the note, and that fades out before the end. It does not use the mod wheel (whose regular vibrato sounds «keyboard-like») but small pitch movements.
- Scoop and fall (sax and brass): attack slightly under pitch at the beginning of a phrase, a small fall of the last long note before a rest.
- Legato: between two close notes of different pitch the first overlaps the second by 18 ms, with no gaps.
- Where the file already has expression (CC11), pitch bend or wheel on the channel, the corresponding intervention is skipped: the file's own prevails.
- With Per-family corrections it works after it and coexists well.
- With Doubling the shadow voice also follows breath and intonation.
- It applies at the next Play. Turn it off to hear the MIDI as written.
External MIDI output
What it is for: sends in real time the same MIDI already improved by this window (cleanup, corrections, humanization, style, transposition...) to a real MIDI device connected to the computer, besides the sound produced internally by FluidSynth. For the program there is no difference between a hardware expander connected by USB, an audio interface with a MIDI port or a virtual synthesizer such as VirtualMIDISynth: they all appear in the same drop-down list.
- The MIDI output menu lists the MIDI devices found on the system; Internal (FluidSynth) (default) sends nothing outside.
- What reaches the device is identical to what FluidSynth hears: same corrections, same key. The original MIDI file is not touched.
- With an external device chosen, FluidSynth's internal sound is silenced: you only hear the expander. For the same reason the timbral reinforcement (doubling with the second synthesizer) is disabled, and the related notice does not appear.
- Stop, Pause and a move on the progress bar reset all notes on the external device, so no stuck sounds remain.
- If the chosen device is no longer connected, it stays selected at the end of the menu as «(not connected)» instead of disappearing: playback continues with the internal audio only.
- Available only on Windows for now; on Linux and macOS the menu lists only Internal (FluidSynth).
- It applies at the next Play: changing device during playback does not move it on the fly.
Audio output
What it is for: chooses which sound card or device olw-player plays on (speakers, headphones, HDMI...). It is separate from the External MIDI output, which is unchanged for expanders.
- Automatic (default): audio always goes to the output active when the program starts and does not move when you plug or unplug an HDMI screen or other devices.
- To use another output pick it from the Audio output menu, just above the MIDI output, and confirm with OK.
- The choice is not saved: at every launch it starts again from the output active at that moment. If the chosen output is gone, the program falls back to the default one and tells you once.
Make piano dynamic jumps more regular
What it is for: corrects the «spring» effect of the piano (instruments 0-7): a note cannot drop by more than 15 velocity points compared with the previous one, so the sound does not jump from loud to very faint.
Fix double hits and abnormal drum panning
What it is for: in «crooked» MIDIs the drums may play all on one side or with every hit doubled (double volume and out-of-phase sound). The option brings the pan back to center beyond ±20, removes the balance (CC8) and reduces double hits (same note, same instant) to one, with the highest velocity. It applies to channel 10 only.
Warn if transposition takes notes out of range
What it is for: changing the key, some notes may leave the range of the MIDI or the instrument. With the box on, the status line warns how many notes are out of range and which instruments (guitars, basses) were raised by an octave. It does not modify the notes, it only informs.
The octave correction for guitars and basses when the key is lowered is instead always automatic: the warning concerns only the message.
↑ indexReal-time sound and mix
How the sound effects work
The effects act on the stereo mix produced by FluidSynth, while it plays, without creating files. Order: tube warmth, multiband bass, bass and treble, low-mids, harmonic brightness, convolution reverb, compression, stereo widening, final limiter. You can change them even while listening with Apply.
Reverb
What it is for: turns on FluidSynth's internal reverb, which gives room and body to the sound. On by default.
If in the main window you chose a room from the Reverb menu (convolution reverb), this internal reverb is turned off automatically: the two do not add up.
Chorus
What it is for: a light doubling with a small pitch variation that gives width to some timbres (strings, pads, guitars). Off by default.
Leveling and peak protection
What it is for: a final limiter (threshold 0.97, release 300 ms) that keeps peaks from distorting, especially after boosted bass and treble. On by default.
Harmonic brightness
What it is for: makes the sound clearer and more open without raising the treble with EQ. It is a harmonic exciter: it takes the part of the sound above about 3 kHz, saturates it slightly to create new harmonics and adds a small amount to the original. The treble level does not change, but cymbals, voice and strings sound sharper. On by default; it replaces the old «Presence on vocal frequencies».
- It has no conflicts with the bass, but together with Treble boost the sound can become harsh: use only one.
- It also boosts the hiss of certain SoundFonts a little: listen again before keeping it.
- Does it apply to the «no reverb» drums? No: only the fixed EQ applies to the separate drums.
Light compression of the mix
What it is for: a compressor on the whole mix (threshold -24 dB, ratio 3:1, attack 20 ms, release 200 ms) that «glues» the instruments and makes them more uniform. Off by default.
Stereo widening
What it is for: widens the stereo image by reinforcing the sides (strength 1.5). Off by default.
Compensated stereo widening: when enabled, a separate worker measures mid and side energy after the other effects. The audio filter reduces the unwanted frequency-dependent gain of 1.5 widening, preserving average spectral balance. Measurement adapts over roughly three seconds of audio and the filter changes gradually; this is not full-song analysis before Play. The callback never waits for the worker or reads files; compensation adds a 64-sample delay (about 1.45 ms at 44.1 kHz). With the option off, the previous path is used. No fixed cymbal cut or changes to manual volumes or MIDI notes.
↑ indexBass boost
What it is for: a +5 dB shelf at 100 Hz to give weight to kick and bass when the SoundFont sounds thin. Off by default.
Low-mid boost
What it is for: a +4 dB bell at 300 Hz that gives body and warmth to guitars, piano, male voices and strings, when they sound «empty». Off by default.
Treble boost
What it is for: a +5 dB shelf at 3 kHz to give brilliance if the SoundFont sounds dull. Off by default.
Full multiband bass
What it is for: gives weight and presence to bass and kick without the usual «boom». The sound below 120 Hz is separated from the rest, made more uniform with light compression (3:1 beyond -24 dB) and summed back with about 45% more. In addition the harmonics of the bass (above 150 Hz) are added: the ear uses them to «rebuild» the low sound, and the bass is heard even on small speakers, phones and cheap headphones, where a simple boost would not be heard. Off by default.
The low band is treated in mono (as is done in mixing for bass and kick): the stereo image does not change.
- With Bass boost it does the same job twice: in the Options window they exclude each other. Choose one.
- On songs with kick and bass already very full (disco, dance) it can weigh down: turn it off.
- Keep Leveling on to avoid peaks.
Tube warmth
What it is for: adds the light saturation typical of a tube amplifier: a touch of even and odd harmonics that makes the sound fuller, softer and less «digital», and gently compresses the peaks. It acts only on the sound below 5 kHz, so the highs stay clean and no hiss is heard. On very quiet signals it does almost nothing: the warmth appears when the sound is full. Off by default.
- With Light compression and Leveling the dynamics are reduced several times: if the song ends up squashed, turn the compression off.
- With Treble boost or Harmonic brightness the added harmonics are brightened and the sound can become harsh.
- On piano and clear voices it may seem an extra color: listen again before keeping it.
- It acts before the convolution reverb, which therefore is not saturated.
Recommended combinations
| Song | Options |
|---|---|
| Default (fine almost always) | All MIDI corrections on, Reverb, Chorus and Leveling on; the rest off |
| Dance, disco, electronic | Turn off Humanization, keep Cleanup and Stereo drums; Electronic style; no bass/treble boost; Chorus off |
| Slow, ballad, jazz | Groove and Compression off; Per-family corrections on (winds and strings); Reverb on |
| Thin «MIDI-like» sound | Harmonic brightness or Treble boost (not both) plus Low-mids; Leveling on |
| Solo piano | Piano dynamics off if the contrasts are intended; Compression off |
| Listening in mono | Stereo widening off; Chorus off |
Defaults, Save as defaults, Cancel, Apply
- Defaults: brings all boxes back to the initial values (the style chosen in the main window stays as it is).
- Save as defaults: uses the current values for all new songs.
- Cancel: closes without changing anything.
- Apply: saves the options for the current song; sound effects change immediately, MIDI options from the next Play.
Contextual audio balance
With Balance enabled and the internal synthesizer, the prepared MIDI is measured using the selected SoundFont and kit. It uses CreaVideoLyrics contextual rules:400ms windows at100ms intervals, constant gains, rhythm compared with corrected accompaniment, and a final foreground wind phrase check. Guide and muted channels are excluded. First preparation can take tens of seconds; verified temporary caches reuse identical measurements. No intermediate WAV files. Playback applies gains through the synthesizer and shadows without changing velocities; processed MIDI export receives equivalent CC7 changes, without applying gains twice during playback. SoundFonts and DSP are not embedded. Measurements are dry before DSP, doubling and live modifiers: OLW requires listening validation separately from the approved CreaVideoLyrics engine1. External MIDI and paths without the required audio API/callback retain prior MIDI balance; hardware audio cannot be measured. A failed required measurement stops playback startup with an error. Turning Balance off retains the previous path.
Final keyboard balance check
With Balance volumes enabled, GM0–7 acoustic/electric pianos and harpsichords are checked against corrected accompaniment, excluding other keyboards from the reference. Foreground phrases use the wind evidence requirements: at least30valid windows,50% coverage,20% above band and one consecutive second of excess;80th percentile ratio,−6dB band and initial cut up to6dB, progressively extended to12dB, retaining stronger previous cuts. No final boosts or fixed cuts; passages without two references are preserved. Family is inferred from the MIDI program, not the sample name. Gain remains constant throughout the song and is included through CC7 in MIDI export when Form settings are enabled. Listening validation is required.
Progressive keyboard cap: the initial cut remains limited to6dB. When measured excess above the band exceeds9dB, only excess beyond9dB is added, up to another6dB (12dB total). The continuous ramp preserves moderate corrections and all timing safeguards. Gain remains constant, not dynamic compression. Wind and other instrument limits are unchanged.
Recovering weak winds
After all corrections, including keyboards, uncut GM56–79 winds are checked for persistent weakness. Gain is proportional, up to6dB toward the family ratio (sax/woodwinds−3dB, brass−9dB) against corrected simultaneous accompaniment energy, excluding other winds. Evidence requires30windows400ms/100ms step, two references,50% coverage,70% below target−0.5dB and one consecutive weak second. Wind windows must be within18dB of peak and above−60LUFS; the80th percentile must not exceed the family guard (sax/woodwinds+3dB, brass−6dB), protecting parts frequently in the foreground. Existing wind cuts are never undone. Gain is constant, preserving velocity and crescendos. For MIDI export, all gains are reduced by the maximum positive gain to preserve relative balance even when CC7 is already127: overall MIDI level is lower, without controller saturation. Listening validation is required; a given song may not meet recovery conditions.
Sax and woodwind recovery
For weak GM64–79 saxes and woodwinds, recovery targets−3dB against corrected accompaniment energy; GM56–63 brass retains−9dB. Maximum recovery remains6dB, proportional to the median deficit. At least70% of comparisons must be below target−0.5dB and previous timing safeguards apply; the80th percentile guard is+3dB for sax/woodwinds and−6dB for brass. GM family is only a role cue, not certain solo recognition. The log includes weak comparisons, consecutive run,80th percentile, target and guard, including zero-gain decisions. A sax median of−8.9dB can receive5.9dB recovery if other evidence conditions hold. Listening validation remains required.