Guide - MIDI → MP3 Conversion

Where to find SoundFonts (and what to know)

The program has no instruments of its own: it takes them from SoundFonts (.sf2, one file with all the instruments) or from .sfz libraries (one instrument per folder, used by the HigherQuality sfz engine). Some free and reliable starting points:

• GeneralUser GS — complete, light and well-made General MIDI SoundFont: a great first global SoundFont. github.com/mrbumpy409/GeneralUser-GS
• Musical Artifacts — large archive of .sf2 and .sfz SoundFonts, each with its license shown on its page. musical-artifacts.com
• SFZ Instruments — free, high-quality .sfz instruments for HigherQuality sfz, for example the Salamander Grand Piano and the VSCO Community Edition orchestra. sfzinstruments.github.io

Be wary of "500 SoundFonts in one pack" collections: very uneven quality and often unknown licenses.

What to know before using them

• License: not every SoundFont can be used in backing tracks you sell or play in public. Always read the license of each file; some only ask you to credit the author, others forbid commercial use.
• Uneven volumes: each SoundFont records its instruments at different levels, and the same MIDI can come out with one instrument much louder than the others. With MaximizeQuality and HigherQuality sfz the program fixes the excesses on its own; with MediumQuality correct them with Vol %.
• Missing instruments: an incomplete or "single instrument" SoundFont may not contain the instrument the MIDI asks for: the program falls back on the global SoundFont, but the sound changes. The grid shows "SF2 without this instrument".
• Very different quality: the same song can sound great or "like a toy keyboard" depending on the SoundFont. Always check with a preview; the automatic search (Options > SoundFonts by family) gives each file a quality score from 0 to 10.
• SFZ without samples: an .sfz is just a list of audio files; if the samples are not in the expected folder the instrument stays silent (the program warns you).
• Memory and time: very large SoundFonts (over 500 MB) slow down loading and audio creation.

Form overview

This is the form where you load a MIDI/KAR, choose how it should be "played" (engine, soundfont, key, volumes) and generate the final audio file (MP3/WAV), which is then used to synchronize the lyrics and create the video.

Choosing the MIDI file: the button next to the file field opens the selection window directly in your Music folder, whatever the system language: on Windows the system Music folder, on Linux the one indicated by the desktop (for example "Musica" or "Music"), on macOS "Music". If it doesn't exist, it starts from the user folder. After choosing a file, the next time the window starts from that file's folder. .mid, .midi and .kar files are shown (uppercase extensions too); with "All files" you see everything. You can also drag the file directly onto the field.

The typical flow is: load MIDI → choose engine and SoundFont → adjust key/volumes/channels → preview to check the result → create the final audio file.

Choosing the engine

The MIDI itself contains no sound: something has to "play" it with virtual instruments. That's the engine's job.

The options

• MediumQuality (FluidSynth, engine 0): plays the whole song in a single pass with the global SoundFont (.sf2); it's the fastest, ideal for tests and previews. From the Configure Channels grid it uses Vol %, Key and Mute (written into the MIDI before playing it), but not the per-channel SoundFonts. It is the recommended engine to start with (see Getting started).
• MaximizeQuality (FluidSynth, engine 1): renders each channel separately with its own .sf2 SoundFont (chosen in the configuration grid) and then mixes them: much better separation and air, somewhat longer times.
• HigherQuality sfz (sfizz, engine 2): uses .sfz libraries per channel; the most refined timbre, with the longest render times.

Key, drum intensity and tempo

Key: transposes the whole song by N semitones up or down, without touching the tempo.

Drum intensity: raises or lowers the relative volume of the drums alone compared to the other instruments.

• Key: values from -24 to +24 semitones (two octaves down or up), default 0 (original key). Recommended to stay between -5 and +5: usually enough to adapt to the voice. Large shifts (beyond +/-7) make the instruments sound unnatural (muddy, dark basses if too far down, harsh timbres if too far up); +12/-12 equals a full octave.
• Drum intensity: values from 0 to 200, default 90 (90 = drum volume unchanged; 45 = half, 180 = double). Recommended between 70 and 120: very low values make the drums almost absent and the song loses drive, very high values make them cover the voice and instruments.

Mute channel / Mute channel 2

Two independent selectors to silence one or two specific MIDI channels (e.g. the guide vocal channel, if the MIDI has one).

Values: each selector ranges from 0 to 16; 0 (shown as "-") means no channel muted, from 1 to 16 indicates the channel to silence. Default: Mute channel = 4, Mute channel 2 = 0 (off). Set a number only when you know which channel contains the guide melody (check it with the single-channel preview); otherwise leave it at 0 so you don't lose a useful instrument.

How it shows up in the channel grid

• What happens: the channel indicated by the 1st and 2nd selector appears in the channel grid with the row colored (purple) and Mute = Yes, already when the MIDI is loaded and every time you move the selector.
• Why: that channel is still silenced in the preview and in the final file; before, the grid could show "No" and it looked like the mute wasn't active.
• Good to know: the selected row stays highlighted in yellow: click another row to see the mute color. The marking is purely visual and does not change the song's saved settings.

Overall volume

Adjusts the overall volume of the entire mix, applied after all the other balancing.

Values: from 1 to 100, default 90. Recommended between 75 and 95. Too low: the song sounds weak compared to other backing tracks and will need to be raised later during playback. Too high (close to 100), especially with Mastering and boosts enabled: increases the risk that peaks get squashed by the limiter or distort.

The final file always goes through a -1 dB limiter: even with high values the peaks never go above zero and there are no clicks in players. Above about 80 the volume grows only a little, because the song is already at its clean maximum level.

Reverb (Impulse Response)

The dropdown menu offers several real "acoustic fingerprints" (Impulse Response): they simulate the sound the song would have if it were recorded in a certain environment (hall, church, studio, etc.), instead of a generic synthetic reverb.

Preview

Preview / Stop preview (buttons in the main form, under the volume/reverb controls): plays back the mix with the current settings without generating the final file, with 30-second forward/back to quickly skip to the parts of the song you're interested in.

Single-channel preview with SF2 engines: if the channel has an .sfz assigned and the chosen engine is MediumQuality or MaximizeQuality, the .sfz is ignored and the channel plays with the global SoundFont (previously a message appeared asking to switch to the sfz engine).

Single-channel preview: audio preparation has a 45-second limit (FluidSynth and sfizz). If it's exceeded the preview stops with a message instead of playing back truncated audio: in that case use a lighter SoundFont on the channel or create the audio file directly. The Stop button interrupts preparation at any time.

Stop: stops the preview at any stage, even during "Mixing and MP3 encoding..." (previously in that stage Stop only took effect once mixing was finished, after several seconds). An interrupted mix leaves no half-finished file.

Create Audio File

Generates the final audio file with all the current settings (engine, per-channel SoundFont, key, volumes, reverb, advanced options).

Configure Channels (per-channel SF2 grid)

It is always visible in the main form, on the right (previously it opened and closed with the "Extra Track Options" button, which no longer exists). It refreshes itself every time you load a MIDI; remember to press Save and Apply to keep the song's changes. A dedicated panel where you assign a specific SoundFont/SFZ to each of the 16 MIDI channels, instead of using a single generic one for the whole song.

What you can do here

• Assign a different .sf2/.sfz file per channel (e.g. a better piano only on the piano channel).
• Mute or adjust the volume of a single channel with Vol %: from 0% to 200%, default 100% (the volume the channel has in the MIDI). It works with all three engines, in the preview and in the final file.
  How to read Vol %: it sets the channel's amplitude, like a mixer fader. Our ears are not linear though: 50% does not sound "half as loud".

  Vol %	Change	How it sounds
  150%	+3.5 dB	a bit more present
  80%	-2 dB	slightly further back
  50%	-6 dB	about two thirds
  30%	-10 dB	about half
  10-15%	-16/-20 dB	almost gone (better to Mute)

  The ratio you set is kept in the final MP3 too. Since the finished song is always brought to the same overall volume, lowering some channels makes the others rise slightly (and vice versa): that's normal. Values above 100% are honoured too: if the channel is already at maximum, the program lowers all the others by the same amount. Recommended between 30% and 150%.
• Raise/lower the key of a single channel with the +/- buttons ("Key" column): each step moves by a full octave (12 semitones), with allowed values -1, 0, +1; default 0. It adds to the overall key. Use it only to move an instrument into a more suitable register (e.g. a bass that plays too high): the melody stays the same, only the pitch changes.
• Remove a row's assignment (right-click on the row → "Remove SF2/SFZ assignment") if you want to go back to the default behavior.
• Reset SoundFont for all channels (as on load) (right-click on the grid): clears the SoundFont column for all channels and fills it again with the family SoundFonts, as when the MIDI is loaded. Volumes, mutes and octaves stay as they are. Useful after trying different SoundFonts.
• Use family SF2 / Use family SFZ (right-click on the channel row): assigns to that single channel, for this song, the file set for its family in Options > SoundFont by family. Only the entry for the type used by the chosen engine appears: "SF2" with engines 0 and 1, "SFZ" with engine 2 (with engine 0 the assignment stays in the grid but isn't used for rendering). Useful to put the family file back after a manual choice. Counts as a manual choice and is saved with the song.
• SoundFont and engine: engine 1 (MaximizeQuality) uses the grid's .sf2 files, engine 2 (HigherQuality sfz) the .sfz files; engine 0 (MediumQuality) shows the SF2 column but uses only the global SoundFont for rendering. The grid fills with the family's SF2 or SFZ column depending on the engine, and manually choosing a file (SF2/SFZ button) only shows the right type. When changing engine, files of the type the new engine doesn't use are replaced with the family file of the right type (if the family doesn't have one, the cell is emptied and the channel uses the global SoundFont). If the replaced file had been chosen by hand, a message lists the channels that changed (old file -> new file): save the song's configuration to make the change permanent.
• "SF2 without this instrument" warning: if a channel uses an .sf2 file that doesn't contain the instrument requested by the MIDI (checked by reading the file's preset list: bank 0 for instruments, bank 128 for drums), the text "WARNING: SF2 without this instrument" appears next to the instrument name. In that case the synthesizer will play a different instrument, usually a piano: choose a complete GM .sf2 or an .sfz for that channel. .sf2 files that contain a single instrument (for example "harp.sf2") don't trigger the warning: the synthesizer uses that single instrument anyway.
• Preview the single isolated channel, to immediately tell if the chosen sound works.
• Regenerate this track: lets you fix a single instrument in a few seconds, without waiting again for all channels to render. It redoes only the selected channel (with the new SoundFont/settings) and quickly recomposes the final mix reusing the other already-rendered channels, without redoing the whole export. It only works after a full export with MaximizeQuality or HigherQuality sfz: details and limits below.

Conveniences and automatic behavior

• Last folder: the SoundFont selection window always opens in the folder of the last file chosen (remembered even after the program is closed).
• SoundFont by family: channels without an assigned SoundFont automatically receive the one set for their instrument family (see Options panel). Manual choices are never overwritten.
• Regenerate this track, how it works:
  • It needs a per-channel export already done: a full "Create Audio File" with MaximizeQuality or HigherQuality sfz, at the same overall key and with the same engine selected. Otherwise a message appears and the MP3 is not touched.
  • When you need to redo the full export: after an export with MediumQuality (it creates the MP3 in a single pass and doesn't save the individual channel tracks), after a full Preview with MaximizeQuality or HigherQuality sfz (it rewrites the saved tracks), after an interrupted export and for exports made with previous versions.
  • It doesn't work on a muted channel (from the two Mute channel selectors or from the grid's Mute column) or on a channel with no notes. If a channel that now plays was muted during the export (so its track is missing), the regeneration stops and tells you which channels are missing.
  • What it uses for the channel: the grid file (if the cell is empty, the global SoundFont), Vol %, Key, drum intensity, overall key and the corrections/humanization active in Options.
  • The program used is decided by the file, not by the selected engine: an .sfz is played with sfizz, an .sf2 with FluidSynth.
  • The final mix is redone as in the full export: tracks of the unmuted channels, same level balancing between channels, compensation of low-recorded SFZ, current Mastering and overall volume; reverb (IR or echo) only with MaximizeQuality, as in the export. The new mix overwrites the MP3 of that export.
  • It has a 10-minute limit for rendering the channel and can be stopped with Stop; if you stop it or it fails, that channel's WAV is deleted and the MP3 is not touched (run the regeneration or the full export again). For songs exported before 23/09 a new full export is needed.
  Good to know: with MediumQuality (engine 0) Regenerate this track isn't available, because that engine doesn't save the individual channel tracks. The program checks this by itself and doesn't touch the MP3.
• Tip: for the instruments that stand out most (piano, strings, guitar/harp) a SoundFont dedicated to that instrument sounds much better than a generic SF2 bank; bass and drums are usually fine with the generic SF2 too.

Song versions

Dropdown menu in the main form, near the channel controls. Despite the name it's not about "versions" of the song in a musical sense, but chooses WHERE the per-channel SoundFont assignments come from:

• Global Configuration (from INI): the default assignments, shared by every song you load (saved in the app's general INI).
• Specific Configuration (<name>_sf2.txt): only appears if, next to the loaded MIDI, a file <songname>_sf2.txt exists - a per-channel SF2 configuration saved specifically for THAT song. Selecting it temporarily overrides the global defaults with that specific one.

Options panel (advanced effects)

Gathers a set of optional enhancements, each with its own switch, applied only when generating with "Enhanced Quality" (FluidSynth engine). You can make them apply only to the current song (checkbox "only for the currently loaded song") instead of every song.

Available options

• MIDI Humanization: introduces small natural variations in timing and intensity (micro-timing/velocity), to make the performance less "mechanical". On by default: turn it off only if the song must stay perfectly on the grid (e.g. electronic/dance). When active it also corrects wind and accordion expression: see Automatic sound corrections.
• Reverb: general switch that turns the environment chosen in the main form on/off (see Reverb).
• Mastering: applies a light final chain (a permissive limiter that only contains the peaks) to give more body and uniformity to the perceived volume. On by default: turn it off only if you hear the mix losing dynamics or "punch".
• Piano-spring fix (with adjustable threshold): only on piano, if a note drops suddenly in volume compared to the previous one (e.g. after an accented chord) it "lifts" it instead of letting it collapse. On by default. The maximum drop threshold ranges from 0 to 40 (tolerated velocity drop), default 15, recommended 10-20. Low values (5-10): kicks in almost always, more uniform but less dynamic sound (0 = kicks in on any drop). High values (25-40): kicks in only on the sharpest jumps, more natural contrast but the "spring" effect can reappear.
• Chorus: slight doubling/detuning of the sound, gives more width to some timbres (e.g. strings, pads). On by default. Turn it off if you prefer a drier, more direct sound, especially on solo instruments.
• Widen the stereo image (with intensity): makes the mix wider stereo-wise. Off by default because it can cause problems if the song is later listened to in mono. Intensity ranges from 0.10 to 3.00, default 1.50; 1.00 = no change (below 1.00 the mix narrows). Recommended 1.20-1.50: noticeable widening without artifacts. High values (2.00-3.00): very pronounced effect, unnatural sound and risk of holes in mono playback. With the Fast Pop style leave this option off: that style's Production already narrows the stereo on its own, and adding manual widening on top gives an inconsistent result.
• Bass boost / Treble boost (with intensity in dB): reinforce the low or high frequencies of the mix respectively. Both off by default. Intensity from 0 to +12 dB, default +5 dB, recommended +2/+5 dB (light, almost transparent reinforcement). High values (8-12 dB): "boomy"/muddy bass or harsh/sibilant treble, and the Mastering limiter kicks in often. Turn them on only if the mix is really weak on the low or high end.
• SoundFont by instrument family... (button): a table where you assign an .sf2 or .sfz file to each General MIDI instrument family (Piano, Guitar, Bass, Strings, Brass, Drums, etc.). When you load a song, every channel without a SoundFont already chosen automatically receives its family's one; manual choices made in the channel grid always take precedence. Leave a family empty to use the global SoundFont.
  Electric piano: programs 4 (Rhodes) and 5 (FM electric piano) have their own family, "Electric piano (4-5)", separate from Piano: this way a Rhodes channel no longer receives the grand piano file. If the family is empty the channel uses the global SoundFont; the automatic search fills it by looking for names like "rhodes", "epiano", "wurlitzer".
  Harp: likewise program 46 (Harp) has the family "Harp (46)", separate from "Strings (40-45, 47)": before, a harp channel could receive a violin or viola. The search looks for "harp", "arpa", "concert harp" (and discards "harpsichord").
  Accordion: programs 21-23 (accordion, harmonica, tango accordion) have the family "Accordion and harmonica (21-23)", separate from "Organ (16-20)": before they received the organ.
  Two columns, SFZ and SF2: each family can have both an .sfz file and an .sf2 file. Which of the two is used is decided by the engine (before you picked an "active" file with a click): the SF2 column with engine 1, the SFZ column with engine 2. With engine 0 (MediumQuality) the grid is filled from the SF2 column, but rendering uses only the global SoundFont. Choose... (or double-click) puts the file in the right column based on its extension; an .sfz with no audio samples is rejected. Remove clears the selected cell (on the Family column it clears both). Search SFZ... and Search SF2... each search only for their own file type, each in its own folder (remembered separately): handy if you keep SFZ and SF2 in different folders. They only fill the empty cells of that column. If instead you keep .sfz and .sf2 in the same folder, that's fine too: choose that folder with Search SFZ... (it only takes the .sfz) and then again with Search SF2... (it only takes the .sf2). Single SF2 for all: choose a single .sf2 (typically a complete GM one like FluidR3 GM) and it's inserted in the SF2 column of every family where it's empty.
  Redo the search from scratch: press Clear grid, then Search in a folder..., check the proposals and press Save. If a channel in the song's grid already has a manually assigned file, it stays: clear the cell or reassign it with SF2/SFZ.
• Why do I hear a piano instead of the instrument? With .sf2 files this happens when the channel's SoundFont doesn't contain the requested program (for example Harp 46, Electric piano 4, Flute 73): it happens with .sf2 files that contain a single instrument, with incomplete GM SoundFonts, or with the instrument in a different bank. In that case the synthesizer falls back to the first available instrument, usually a piano. The notes stay correct, only the timbre changes. Solution: use a complete GM .sf2 (for example FluidR3 GM) on the channel or, better, an .sfz specific to that instrument.
• Clear grid (button in the table): clears all families at once, after a confirmation. Used to redo the automatic search from scratch, which only fills empty families. The change becomes permanent only by pressing Save; with Cancel everything goes back as before. It doesn't touch files manually assigned to channels in the song's grid.
• Search in a folder... (button in the table): choose a folder (for example the one where you keep all your SoundFonts) and the program scans it together with all its subfolders looking for .sfz and .sf2 files. For every family still empty it proposes the file that seems most suitable, based on the words contained in the file and folder names (for example "piano", "guitar", "strings", "violin", "brass", "drums", also in Italian: "chitarra", "archi", "batteria"...). .sfz files are preferred over .sf2; for Piano the acoustic piano is preferred over the electric one. Families already filled in are never touched and hidden folders are skipped. Quality-based choice: the name is only used to find candidates; the program then opens every candidate (up to 25 per family) and measures, only for the instrument of interest: how many different samples cover the keyboard, how many dynamic layers it has (different soft/loud sounds), the sample rate (44.1/48 kHz or less), the bit depth (16 or 24), whether it's stereo and how large the audio data is; for .sfz files also the round robin alternations. This produces a score from 0 to 10 and the highest score wins (in a tie, the most relevant name). For .sfz files all the referenced audio files (wav/flac/ogg) are checked: if more than 20% are missing the file is discarded, if only a few are missing it stays in the running but loses 3 points. An .sf2 with multiple instruments that does not contain the family's instrument is discarded (it would play a different instrument). If no file has a suitable name for a family, with Search SF2... a complete GM SF2 (at least 100 instruments) present in the folder that contains that instrument with a score of at least 3 is proposed. At the end a message lists the proposals with the reason, for example "Piano -> Salamander.sfz, quality 10/10: 30 samples across the keyboard, 16 dynamic layers, 48 kHz, 24 bit, stereo [the best of 6 candidates]", and the families for which nothing was found. For reference, FluidR3 GM scores 1-5. The summary appears in a wide, scrollable window and is also saved in the user's home folder, to check later: CreaVideoLyrics_ricerca_SFZ.log (Search SFZ...) and CreaVideoLyrics_ricerca_SF2.log (Search SF2...), each with date, folder scanned and the last search performed. Libraries without .sfz: during Search SFZ..., if it finds an audio sample library (wav/flac/ogg) that has no .sfz file, the program creates one itself in the samples folder, named <folder>_auto.sfz, and puts it in the running with the others (the summary shows "[.sfz created automatically]" and the list of files created). At least 3 different notes readable from the file names are needed ("C4", "a#3", "Db2", "060"; using the convention C4 = middle C = 60); dynamics come from the name or subfolder (p, mf, f, ff, soft/hard, v1, v2...) and alternations from "rr1", "rr2"... Files with noise, release, fx, gliss, fret, mute in the name are ignored. Nothing is ever created for the "advanced" samples inside a library that already has its own .sfz files (noises, releases, articulations the author didn't use). The created file can be edited or deleted. The score measures the richness of the samples, not taste: always check the result (you can fix a row with Choose... or Remove) and then press Save to confirm; Cancel discards all proposals. Variants that require keyswitch or mod wheel are disadvantaged (names with "KS", "keyswitch", "modwheel", "DXF", "CC1", "xfade"); if an .sfz plays silently, the most common cause is that its audio samples aren't found in the expected folder. The check also reads paths written with variables (#define $NAME ...) used by many orchestral libraries. The same check applies when you manually choose an .sfz for a channel in the grid (SF2/SFZ button): if samples are missing the SFZ without audio samples warning appears and the file isn't assigned, because it would play silently. If the entire preview mix turns out silent (for example the only active channel has an SFZ with no samples) volume normalization is skipped instead of blocking the preview with an FFmpeg error: in that case check the channels' SoundFonts. The chosen folder is remembered for next time. For very large folders the search stops at 20000 files: in that case choose a more specific folder.

Recommended settings by genre

Values found by comparing our renders with the original recordings (volume, timbre across frequency bands and stereo width measured). They're a starting point: they're saved for a single song with "Use these options only for the currently loaded song".

Genre preset: in the Options panel, bottom left, the Genre preset menu with the Apply preset button sets all the items below in one go (including engine idx 1 and the main form's IR reverb, applied when you press Save or Apply and Listen). After the preset every item stays editable by hand. Available presets: Dance / Disco / Funk (IR "Ruby Room") and Slow / Ballad / Jazz (IR "Hall", wind echo attenuated) and Pop (IR "Studio", short and dry, narrow stereo) and Blues (IR "Ruby Room", club room, wind echo attenuated; for Slow Blues and Blues Shuffle) and Rock blues (IR "Studio", corrections with groove and glue, full winds). The presets only adjust the sound: to transform the song (instruments, dynamics, production) there's the Style Choice... button.

Dance, disco, funk (many instruments, rhythm up front)

• Engine: idx 1 (per-channel rendering): much better separation and air than engine 0.
• FluidSynth Chorus: off. It doubles and dephases every instrument: instruments "smear" across the whole space and you can no longer tell where they are.
• Stereo widening: off (1.2 at most). With high values the sides become as loud as the center and the left/right distinction is lost.
• Bass and treble boost: off. Disco kick and bass are already full: the boost muddies the low end.
• Corrections by family: all on, especially Groove for mechanical channels, Mastering: presence and Mastering: mix glue.
• IR reverb: optional; "Ruby Room" or "Studio". It doesn't receive the bass, so it doesn't muddy the kick and bass.

Slow songs, ballads, jazz (few instruments, winds and piano in the spotlight)

• Engine: idx 1.
• IR reverb: on ("Hall" or "Ruby Room"): here the room is part of the sound.
• Corrections by family: winds, strings and voices on (legato, sax vibrato, slides). Groove isn't needed (slow songs are already played), mix glue is better off so as not to remove dynamics; presence to try out, on piano and voice it can come out harsh.
• Wind echo copies: if a sax is echoed on other channels and sounds fake, choose "Attenuate" or "Remove".

Style Choice

The Style Choice... button (on the right, under "General Options") transforms the loaded song into a style, always in the same way. Unlike the genre presets in Options, which only adjust rendering, here the song itself changes, on three levels that can be switched on and off:

• Arrangement: which instruments play (e.g. the piano becomes an acoustic guitar with strummed chords);
• Interpretation: how they're played (more restrained, softer dynamics, ritardando on the last bars);
• Production: the final sound (equalization, compression, reverb, fade-out), applied in the preview and in the MP3.

The Intensity slider dials all three (70% = reference values). Apply creates a new MIDI <song>_<style>.mid in the _Files folder, with its own configuration (copied from the song's; channels that change instrument receive the new family's SoundFont) and loads it: the original isn't touched and the MP3 has a distinct name. Then you listen with the normal preview. This MIDI only contains the style transformation: for a MIDI that also includes corrections and humanization, to use outside the program, there's Export corrected MIDI. Reopening Style Choice from a styled version, the choices restart from the ones used and the style is always redone starting from the original (never one style on top of another); Back to original reloads the starting song.

Production that adapts to the song: each style has a reference profile measured on a reference version of the genre: how much weight the bass, mids and treble carry (7 bands), how wide the sides are compared to the center and how much dynamics there is. During the mix the program measures your song, with the SoundFont and engine you're using, and works out on its own the equalization, stereo width and compression to get closer to the profile; a second measurement corrects the remaining error. The result therefore no longer depends on the sound bank. The Intensity slider says how close to get (70% = up to the reference). The final MP3 message shows the remaining gap ("style profile: max gap ... dB"). Mixing a styled song takes a few seconds longer.

Available styles: Soft acoustic Ballad, tuned by listening to and measuring "Dove": acoustic piano → steel acoustic guitar, electric piano → nylon guitar (slower, softer strumming) (downstroke on the beats, lighter upstroke on the offbeats, E2-G5 voicing with at most 6 strings), velocities brought closer to the channel median, ritardando up to +15% on the last two bars, bass +3 dB, less 600 Hz, 4.5 kHz and 8 kHz, soft compression (dynamic range from about 7 to about 4 LU), IR reverb +20%, 6-second fade-out.

Fast Pop, tuned by listening to and measuring "Ora che non ho piu' te": designed for guide MIDIs already arranged in pop style (bass, rhythmic guitar/synth, real drums) - Arrangement doesn't change instruments or notes. Interpretation: if enabled, a dynamics crescendo towards the choruses (up to 25% more velocity at full intensity), automatically detecting verses and choruses from the MIDI; drums and strings/pads are however trimmed in the choruses in proportion to the boost given to the other sounds, so they don't end up disproportionate. Production: reinforced brightness/presence (up to +11 dB 6-12 kHz, +8 dB 2-6 kHz, +6 dB 12-16 kHz, +4 dB sub/bass), faster and more decisive compression, short reverb, stereo slightly narrowed towards the center.

Slow Blues, tuned by listening to and measuring "Tennessee Whiskey" (Chris Stapleton): for slow blues and soul-blues (12/8 or slow 4/4) with guide MIDIs already arranged. Arrangement: acoustic guitars → clean electric guitar (the program discards distorted/overdrive SoundFonts and falls back to the global SoundFont), organs and synth pads → Hammond organ, pick/fretless/slap/synth electric bass → finger bass; piano, winds and melody stay as they are. Interpretation: no change to notes or rhythm (no swing added), drums a bit lighter, slight crescendo in the choruses (+15% at intensity 70, with headroom: channels already close to 127 are rescaled instead of clipping), final ritardando only if the MIDI doesn't already have one. Production: cleaner low-mids, more presence and a touch of brightness, light compression and peak limiter (less jutting hits, like in professional mixes), no final fade-out (the song is sung right to the end). Pair it with the Blues genre preset in Options.

Blues Shuffle, tuned on "Sweet Home Chicago" (Blues Brothers): for blues with a "skipping" feel (triplet eighths) and a band with horns. Arrangement as in Slow Blues. Interpretation: no note changes; the program checks whether the MIDI is already shuffled and says so in the summary (if it has straight eighths, for now it leaves them straight); slight crescendo in the choruses; full drums. Production: less bass, much more presence and brightness (horns and guitars up front), wider stereo, peak limiter, no fade-out.

Rock blues, tuned on "Diavolo in me" (Zucchero): for rock-blues with straight eighths or sixteenths, distorted guitars and horns. Arrangement: no change. Interpretation: crescendo in the choruses more decisive than in the blues styles (+20% at intensity 70), with headroom on already-high velocities. Production: more presence (guitars and horns up front), a touch less air, stereo slightly narrower, peak limiter, no fade-out.

More styles will be added the same way.

Opening a MIDI with double-click or "Open with"

• What it does: Crea Video Lyrics appears among the suggested programs for .mid, .midi and .kar files (right-click on the file → "Open with"). Opening a MIDI this way, after the license check, opens "Midi to Mp3" directly with the file already loaded.
• Windows: registered by the installer, without changing the default program. To always use it with a double-click: right-click → Open with → Choose another app → Crea Video Lyrics → "Always use this app".
• macOS: the Finder offers it in "Open With" after the app's first launch; for double-click: Get Info → Open with → Crea Video Lyrics → "Change All".
• Linux: the AppImage declares the MIDI types; the "Open with" menu shows it if the AppImage is integrated into the system (for example with AppImageLauncher or Gear Lever).

Export corrected MIDI

• What it does: the Export corrected MIDI button (under "Create Audio File") saves the file SongName_corretto.mid next to the loaded MIDI, with every channel passed through the same rendering chain (the _corretto suffix is the same in every language). The original is never modified. If a styled version is loaded, the file goes into the _Files folder with the version's name (e.g. SongName_<style>_corretto.mid).
• Difference from Style Choice: the MIDI created by Style Choice only contains the transformation of the song (arrangement, interpretation); corrections by family, humanization, key, volumes and pan are applied "on the fly" during preview and rendering, and stay out of the file. The corrected MIDI instead writes them all in: it's the file to take outside the program. They can be used together: apply a style, then export the corrected MIDI of the styled version.
• Corrections by family: legato, vibrato, expression, tails and releases, repeated notes (see "Automatic sound corrections").
• Humanization: dynamics (velocity) recalibrated and varied, wind breathing, piano pedal, softened note-offs, pan. Includes the corrections too. Note timing (attacks) is never shifted.
• Form settings: key, per-channel octaves, volume of each channel (written as CC7) and attenuation of doubled instruments.
• Exclude muted channels (off by default): the exported MIDI contains every channel, melody included. Turning it on gives a karaoke MIDI, without the muted channels, just like the audio.

When to disable the options: turn off "Form settings" to get the complete song in its original key; turn off humanization for a MIDI that's just corrected, to be refined by hand in a sequencer. IR reverb, mastering, stereo widening and boosts don't go into the MIDI (they're audio effects), nor does the string pad.

Automatic sound corrections

Three corrections the program applies on its own during preview, single-channel preview, "Create Audio File" and "Regenerate this track". None of them modify the original MIDI: it always works on temporary copies.

1. Wind and accordion expression

• Who it applies to: wind channels (sax and reeds, brass, flutes: GM programs 56-79) and accordion / harmonica (21-23), with all engines (MediumQuality, MaximizeQuality and HigherQuality sfz), including the fast MediumQuality preview.
• What it does: legato phrasing (winds only): if there's a gap shorter than 90 ticks between two notes, about a sixteenth, the first is extended up to the second, and isolated notes shorter than 80 ticks are extended to 80 ticks; attacks never change, so rhythm and notes stay as written, but the sax no longer sounds "staccato like a keyboard"; re-centered vibrato: if during a held note the vibrato only oscillates below (or only above) the pitch in effect at the attack, it's shifted to oscillate around the note and narrowed to 40% of its width (a vibrato almost a semitone wide sounds out of tune; a real sax's sits around ±15-20 cents), so the instrument no longer sounds flat or "wobbly"; more gradual volume descents (expression no longer drops by more than 25 units every eighth or so, so drops resemble a breath rather than a step); very short expression holes removed (CC11 volume that crashes by more than 40 and comes back up within 60 ticks, about a sixteenth); compressed expression between 50 and 127 instead of 0 to 127.

When it applies / how to disable it: it's active when MIDI Humanization is active in Options. Disabling Humanization plays the MIDI exactly as written (useful if the MIDI is already tailored for a specific virtual instrument that handles bend and expression well).

Values: minimum expression 50 out of 127 (about -8 dB below maximum, instead of silence).

Pitch: pitch bend is never reduced or zeroed. The only intervention is re-centering the vibrato, which only kicks in when all these conditions are true: the note is held, after the first 120 ticks (about an eighth, the attack with its "slide" stays intact) there are at least 4 bend values that go up and down at least 4 times (a vibrato, not an intentional bend in one direction), and their average is more than about 1/13 of a semitone away from the initial pitch on the same side. The check is done note by note even when legato phrasing makes notes overlap or touch: when a new note starts, the previous one's vibrato is evaluated and closed there. At the end of the note the bend returns to the original value. The rest of the bend stays as written. In many MIDIs the bend is part of the pitch itself: for example it rises two semitones and stays up for the following notes, written lower. Reducing or zeroing it would make the instrument out of tune.

2. Compensation for low-recorded SFZ files

• Who it applies to: every channel using an .sfz file (HigherQuality sfz engine).
• What it does: reads the volume= values of the file's regions and averages them; if the average is below -3 dB it raises that channel to bring it back towards 0 dB.

Values: maximum correction +12 dB; .sfz files with an average between -3 and 0 dB (or without volume=) don't change. The correction adds to the channel grid's Vol %: if an instrument still seems too quiet or too loud, adjust it there.

2b. SFZ with wrong transposition

• Who it applies to: channels with an .sfz file, with the HigherQuality sfz engine, excluding drums (channel 10): in drum kits each note corresponds to a different sound (kick, snare, cymbals...), so shifting them would change the instruments or silence them.
• What it does: if every region of the file has the same transpose= of 12 or more semitones (up or down), that channel's notes are shifted by the same amount in the opposite direction, so the instrument plays at the pitch written in the MIDI. The channel grid shows "SFZ transposed by N semitones (compensated)".

Good to know: different transpositions between regions, or ones smaller than 12 semitones, are considered intentional and are not touched. With SF2 engines the .sfz file isn't used, so the compensation isn't needed.

2c. Final mix volume (normalization)

• Who it applies to: the complete preview and final file mix, with all engines.
• What it does: measures the perceived volume of the entire song and applies a single fixed gain, the same from start to finish, to bring it to about -16 LUFS; peaks that would exceed -1 dBTP are contained by a fast limiter (up to 3 dB of limiting, like in a light mastering pass: more would squash the drum hits in the choruses; beyond that, the gain is reduced).

Speed: the mix of all channels, with effects and reverb, is now computed only once (previously it was redone for every measurement pass and for the MP3) and at 48 kHz, the MP3's rate. The volume measurement also now uses a filter about 7 times faster (ebur128, same values). The "Mixing and final Mastering" stage now takes about half or less (in tests: from 47 to about 20 seconds for a 4-minute song, with mix Glue from 72 to about 22); the sound doesn't change. At the end of conversion the message shows the time of the final stage, split between channel balancing, mix with effects and measurement, glue measurement and MP3 creation: useful to see where time is spent. Some free disk space next to the MP3 is needed for a temporary file (about 90 MB per 4 minutes of song), which is deleted at the end. The progress bar now grows throughout the whole conversion: 10% MIDI preparation, 20 to 80% one step per rendered channel, 82% channel balancing, 88 to 100% mix and MP3 creation.

Values: target -16 LUFS (typical of streaming services); maximum peak -1 dBTP; correction limited to ±20 dB. The form's Overall volume is applied after normalization, so it's a real output volume control: 100 = normalized level (-16 LUFS), below 100 quieter, above 100 louder (in that case a limiter prevents peaks from saturating).

2e. Corrections by instrument family

• Who it applies to: every channel except drums, with Humanization active, in preview and final file.
• What it does: recognizes each channel's instrument family and applies only the corrections suited to how that instrument is played:
  • Solo winds and synths (GM 56-87): legato under 90 ticks, isolated notes at least 80 ticks, re-centered vibrato, soft expression (CC11 from 50 to 127).
  • Sustained strings (GM 40-44, 48-51; pizzicato excluded) and voices/choir (52-54): lighter legato (under 60 ticks), re-centered vibrato, soft expression (CC11 from 40 to 127).
  • Accordion and harmonica (21-23): re-centered vibrato and soft expression, no legato.
  • Bass (32-39): very short notes reach at least 40 ticks, without touching the following note: the staccato stays, the "cut off" disappears.
  • Guitars (24-31): minimum tail of 30 ticks; bends landed on target: if a bend (pitch bend) stays close to a semitone or a tone but doesn't reach it (within 35 cents, for example +175 instead of +200), it's scaled so it lands exactly on the note; if the bent note has a wide vibrato written (oscillations of more than a fifth of the bend, typical of MIDIs: down to almost a semitone below the note), its depth drops to 40%, always below the note as a guitarist does. Vibrato on a non-bent note (the pitch rises and returns to the note, under a semitone) if wider than 40 cents is halved: a vibrato from 0 to +70 cents, entirely above the note, sounds sharp. Small bends (under 65 cents) and glissandos that don't settle on a note stay as they are. The pitch bend range is the one set by the MIDI (RPN), otherwise 2 semitones. On overdrive and distorted guitars (29-30) played as a solo line, if a note starts while the previous one is still sounding, the previous one is closed at the new note's attack: with distortion even a few tens of milliseconds of two close notes (a tone or semitone apart) together make a dirty sound. Double stops and chords (notes starting together) stay intact. Also on these guitars, double grace notes are merged: the same note split into two with a very short gap (up to about 1/12 of a quarter), when the restart happens inside a bend (a bent string doesn't get replucked), when the gap is nearly zero, or when the first note is very short. Normally repeated plucked notes stay as they are. Fast legato passages (27/09): chains of very short notes (up to 160 ms) attached to one another and 1-2 semitones apart (e.g. 43-44-45) are no longer plucked note by note: they become a single note with the steps done by pitch bend, like a guitarist's hammer-on / pull-off. Only when in the passage the guitar plays one note at a time and with no bend in progress. The same applies to bass (32-39), with the same thresholds. Multi-voice clusters (very short double or triple stops that slide a semitone or two up to a chord or a long note, typical of keyboard-played MIDIs) are instead removed: the arrival chord or note starts in place of the first attack. Keyboard-style slides (27/09): when the MIDI brings the guitar into a chord by sliding up from below with the pitch bend wheel (or while the previous note is still ringing), you hear everything drop and rise back; these attacks become straight, with a fast vibrato at the end of long notes to accompany the change. The slide preparation done on a still-sounding note becomes a jump on the attack, and modulation-wheel vibrato (CC1) on guitars is reduced to about a third. The sustain pedal (CC64) on guitars is also removed: it's a keyboard gesture that keeps notes sounding past their end, and with bends it dragged them along. Held bends and slides onto a plucked note stay as they are.
  • Pads (88-95): release extended up to 40 ticks, within the silence before the next note.
  • All (drums excluded): overlapping repeated notes. If a note restarts while the same note is still held, the first is closed at the second's attack.

Values: the ticks shown refer to 480 PPQ and are rescaled to the MIDI's PPQ. Attacks (note starts) never change, so rhythm and notes stay as written. To disable all corrections, turn off Humanization in the Options panel.

2f. Dynamics and doublings

• Dynamics: humanization recalibrates each channel's velocity towards the 25-127 range, but now only halfway: flat dynamics open up, without sudden volume drops in quietly-played passages any more. On winds, solo synths, sustained strings and voices, the written velocity stays the original one (dynamics come from CC11 expression instead), with only a small random variation.
• Tight doublings: when two identical channels (same notes, same instrument) start less than a thirty-second apart (about 40 ms) they don't make an echo but a doubling that muddies things: the second is lowered by about 7 dB, as well as being separated in stereo.
• Very wide wind bends (doit, jazz fall): if during a note the pitch bend moves away by more than 3/4 of a semitone and then returns to the note, the bend is reduced to 60%. Bends that stay raised (used as pitch for the following notes) are not touched.
• Wind vibrato: vibrato written with pitch bend (typical of a keyboard's modulation wheel: slow, up to 3/4 of a semitone wide, regular and present right away) is removed and replaced with a sax-style jaw vibrato: it starts after about a third of the note (only on notes of at least about half a second), builds over two oscillations, about 5.5 per second, ±12 cents, slightly irregular in amplitude and speed. At the end of the note the pitch returns exactly.
• Fast wind passages: in notes less than an eighth apart the velocity gets closer to the previous note's (no attack jumps from one note to the next) and each note overlaps the next by a hair. On a sax, the tongue strikes the first note and the rest are slurred: the passage "blends" instead of sounding clean-cut as on a keyboard.
• Chromatic passing notes on winds: one or two very short grace notes (less than a sixteenth) a semitone apart leading to the real note (e.g. F, F sharp -> G) are the typical ornament played on a keyboard: on a real sax it's a lip slide. If the grace notes are below the real note (rising towards it) they become a smooth pitch bend, the saxophonist's "scoop", and the real note starts right in tune immediately. If they're above (dropping towards it, for example C sharp -> C after a jump) they're removed: a fall from above onto the attack, with the main note already there, is not something a saxophonist does and sounds fake. The real note starts straight at the moment of the ornament. Chromatic runs of three or more notes stay as they were.
• Pan of doubled instruments: the pan written in the MIDI on those channels is ignored, so the stereo separation stays throughout the song.

2g. Choosing corrections (Options > Corrections by family)

• What it does: the Corrections by family... button in the Options panel lets you turn each group's corrections on or off: solo winds and synths, sustained strings, voices, accordion, bass, guitars, pads, the repeated-notes correction and groove for mechanical channels. Applies to every song, or, with "only for this song", to the loaded song.
• Wind echo copies: when a wind instrument is repeated on other channels with a delay (an echo effect written in the MIDI) you can choose Keep (as in the MIDI), Attenuate (about -9 dB) or Remove (only the main wind stays). Applies to preview, audio file and MIDI export (with "form settings").
• Simultaneous doublings (on by default): if a channel plays at least 80% of the same notes as another of the same instrument family at the same instant (unison or one/two octaves apart), for example two guitars copying each other note for note, the two volumes add up and the part ends up too loud. The copy is lowered by about -7 dB; the channel with more notes stays full (in a tie, the piano). Different from echo: here the notes start together, not delayed. Turn it off if the doubling is intentional and must stay full (for example a unison used as an effect).
• Secondary pianos at the level of the main one (on by default): among piano channels (GM 0-7) the main one is the one with the most notes. Another piano whose typical (median) velocity is more than 10 points above the main one's is lowered as if it were playing at the main one's velocity (minimum factor 0.3). Example: a secondary piano at 96 against a main one at 77 = about -4 dB. Needed when an intro or reinforcement piano is written much louder than the song's piano and "starts blaring". Turn it off if the second piano really must stand out (for example a solo).

When to disable: if a family sounds better as written (e.g. guitars already well played, strings already with good legato). Default: all on, echo kept.

2h. Groove for mechanical channels (disco, dance)

• Who it applies to: "machine-played" channels, meaning at least 85% of notes at the same velocity (at least 16 notes); on drums the check is done per piece (kick, snare, hi-hat...). Already-played channels aren't touched. With Humanization active and the "Groove" item on in Corrections by family.
• What it does: accents based on position in the bar (downbeat full, upbeat a bit lighter, sixteenths lighter, first beat slightly stronger); snare about 7 ms behind the beat, hi-hat on the upbeat 3-4 ms behind, kick always on time; open hi-hat on the upbeat emphasized; octave bass with the higher note lighter; ±3 ms and ±4 velocity randomization on the other mechanical instruments; dynamics that "breathe" by ±5% every 4 bars.
• Fills: when snare and toms are dense (at least 4 hits in a quarter note) the position accents give way to a crescendo towards the next beat, so the push into the chorus drives forward instead of dropping off. In "launches" (snare alone, without kick or hi-hat, usually together with wind and string stabs) the hit is full (+15%), and crashes get a full accent (+30%): in MIDIs they're often written weak and disappeared under the stabs.

When to disable: for electronic songs where the machine's precision is intentional (techno, electro). Humanization is reproducible here too: the same song exported twice gives the same MIDI.

2i. Automatic channel balancing

• What it does: in engines that render channel by channel (MaximizeQuality, HigherQuality sfz) every channel is measured in perceived loudness (LUFS) and those well above the others (more than 6 dB above the middle value) are brought back in line. Drums included. The measurement is always made as if the channel were at 100%: your Vol % and the Drum intensity are applied afterwards and are never cancelled.
• Example: in "Dove" the synth brass of the airfont SoundFont played about 12 dB above every other instrument: now it is brought back to the level of the loudest ones (piano, drums), and from there you adjust it with Vol %.
• Repeatable tests: humanization (small timing and dynamics variations) always uses the same sequence for the same song: with the same settings preview and final file sound identical, and two tests differ only by what you changed.
• Speed: channels are measured in parallel, one per processor core (on 9 channels from about 35 to about 8 seconds).

2i-bis. Skewed or doubled drums

• What it does: on channel 10 (drums), always, even with Humanization off: ignores balance (CC8), which General MIDI drums don't use; recenters a pan (CC10) shifted more than 20 from center (0-127, center 64); reduces to one any double hits (same piece at the same instant), keeping the higher velocity. Applies to rendering, preview, Regenerate track and Export MIDI.
• When it's needed: MIDIs where two drum tracks (for example one all-left and one all-right) were merged into a single channel: the drums ended up all on one side and every hit sounded twice. After the correction, kick and snare go back to the center and hi-hat, toms and cymbals spread out as in the SoundFont's kit.
• Note: if the hits were previously doubled, the drums now play about 6 dB quieter (they were doubled): if needed, raise them with channel 10's Vol% in the grid.

2j. Mastering: presence and mix glue

• Presence (Options > Corrections by family): tonal balance of the whole mix: bass below 100 Hz -3 dB, +4 dB with a wide bell at 2.5 kHz, air +4 dB above 8 kHz. Brings guitars, clavinet, winds and strings forward (the instruments that give "grit" and that sit at the sides on records) and lightens the low end.
• Mix glue: compressor on the already-normalized mix (threshold -24 dBFS, ratio 3:1, attack 20 ms so the kick passes through, release 200 ms, soft knee, +3 dB makeup gain), then the song is measured again and brought back to -16 LUFS (one extra measurement pass), with the safety limiter. Reduces the distance between the peaks and the rest: the song sounds more compact and "all together", like a record.
• FluidSynth Chorus and stereo widening: in the comparison with the original, they were the ones "smearing" instruments across the whole space (sides as loud as the center): turning them off brought the stereo scene back almost identical to the original. In songs with many instruments it's better to leave them off.
• IR reverb without bass: only the part of the mix above 250 Hz reaches the reverb. Kick and bass stay dry and centered, and the reverb gives air to the instruments without muddying the low end.

When to disable: in acoustic slow songs and piano-and-voice pieces, presence can come out harsh and glue can remove dynamics: try with and without. Default: off. With glue enabled it's recommended to leave the bass boost off.

2d. Natural closing of final notes

• Who it applies to: every channel, with every engine, in preview and final file.
• What it does: when a channel is prepared for rendering, at the end of the track any notes still open are closed with their natural release and a 4-beat tail is added before the end of the file.