Optimizing Figures

Make the most of your recorded or imported figures by optimizing them for best performance in any harmonic context.

Re-Usability is Top Priority

Figure recognition is very ambiguous and highly depends on harmonic context (which is also ambiguous). Ambiguous here means: There are so many - billions - possible results that it's extremely hard to decide which is best. Synfire's AI goes to great lengths to make this decision. Still, the results can never be perfect. 

Even more so as "perfect" is also relative: A figure that seems perfect at first, in that it faithfully preserves the original performance, may not work well in other harmonic contexts. This unlimited re-usability however, should be your top priority goal for figure optimization. A figure is best, if it works in any song.

Note: Static pitch symbols, e.g. for drums, do not need optimization. This tutorial only applies to dynamic figures.

The Checklist

This checklist provides an overview of what to look for and how to fix it. All aspects mentioned here are discussed in detail below, in the following paragraphs.

  1. Are chords, bass and melodies cleanly separated and symbol types used as expected?
    (=> pre-tag notes in take, transpose parts or entire take, try other presets)
  2. Does segment grouping make sense?
    (=> re-group in figure, pre-group in take, avoid clusters of lone symbols)
  3. Are there anchors that should be changed?
    (=> set different anchors)
  4. Are there bass movements that should be preserved?
    (=> group bass symbols, keep segments short)
  5. Too many bass symbols?
    (=> check playing range, transpose take, thin out take)
  6. Any narrow, parallel lines that need grouping together?
    (=> group narrow lines in single segment to avoid them clashing)
  7. Any chords that may sound better with abstraction or the Full Chords option?
    (=> use abstraction, pre-tag as purple symbols, flag figure segment as Full Chords)
  8. Does the use of accidentals make sense?
    (=> disable or remove accidentals)

Tip: Always imagine the figure was used in a totally different harmonic context. Where are potential weak points that need fixing?

As figure recognition technology will make progress and evolve over time, more and more of these these optimizations may be automated and taken care of by Synfire in the future. For now, you might need to look at a figure and fix it yourself, if necessary, in order to get great re-usable figures that work everywhere. 



Selecting a Preset

Figure recognition presets recall different strategies and settings for typical musical instruments and patterns. The default preset detected by Synfire when you select "Auto-Detect" is usually a good start. If the resulting figure does not meet your expectations, you should try a different preset. Only after this also does not work for you, you can try and tweak the individual settings.

How a Perfect Figure Looks Like

Of course, in order to optimize a figure, you need to have an idea how a figure actually should look like. The sections below discuss various aspects of a figure and how to recognize and fix them.

Symbol Type

After figure recognition has run, bass notes should convert to bass symbols (magenta), chord notes to chord symbols (purple or green), melodic lines to vertical or horizontal scale symbols (turquoise, blue), etc. Here's an example of the same take, converted to a good figure and a bad one (both were run with different settings):


Often it's difficult to determine the best symbol type in polyphonic material, even for a human, especially where melodies, chords and bass occur all in the same take. If you feel Synfire made a mistake ...

  • You can force Synfire to use a certain symbol type by pre-tagging notes in the take: Select notes and double-click on the symbol type they are supposed to convert to. This is helpful where Synfire fails to separate bass notes from the rest, for example.
  • You can pre-group segments in the take already: Select the notes in the take and group them as you would do with a figure.
  • Check if you have the desired symbol type enabled in the first place.
  • Try a different preset or algorithm setting.

So if you already know how you want the final figure to look like, you can pre-tag and/or pre-group notes in the take, before you apply the figure recognition (select notes in the take, then group them or double-click on symbol type to tag them).

Segment Grouping

The grouping of two or more symbols as a figure segment ensures the melodic movement between them is preserved, regardless in which harmonic context the figure is rendered.


You need to decide how to split a long melody into smaller segments (or whether the decision Synfire made is ok). In theory, it's quite possible to group a melody spanning multiple measures, even an entire container. While this would indeed preserve the relative movement, it would dramatically widen the pitch range due to voice leading constraints.

Splitting up a melody into segments is a creative decision to which there is no single answer. These hints may help you:

  • Space: Divide a melody at longer gaps.
  • Rhythm: Divide a melody where rhythmic patterns repeat, change or come to a rest.
  • Symmetry: Arpeggios, broken chords, symmetric runs.
  • Lyrics: Divide a melody where sentences and longer words end.

Synfire usually makes good decisions, but you should have a look, if you feel the result doesn't quite sound as expected.

Tip: Avoid clusters of lone symbols. Always consider grouping them where other symbols are near.

Pattern Recognition Challenges

The example take below demonstrates the challenge of distinguishing melodies from chords. While the bass seems obvious and melodic lines and a couple chords are easy to spot, the question remains which notes to include with a chord and which notes of a chord actually belong to a melody (they can't be included in both).

In the example below, it might be most appropriate to pre-group all lines as melodies, except for the bass. Synfire defaults to sustained chords with melodic ornaments for this one, which is not ideal.


Narrow Melodies

Dealing with parallel melodies in the same figure requires special attention, especially if they are placed close to each other (narrow voicing).

In the left example, the rendered output of the two lines may touch or cross. In order to maintan the intended distance between the lines, the right example groups both lines in a single segment. Always keep in mind that putting symbols in one segment does still allow them to overlap and belong to multiple voices.


Odd Positions

of_zigzag2.pngDon't pay too much attention to the shape or sequence of connecting lines inside one segment (see image to the right). A segment is merely a group of symbols ordered by time. The individual notes of chords and vertical clusters, for example, may be playing at slightly different times to preserve a human touch. The connecting lines may look a bit zig-zagged, which is ok. 

The segment in the picture to the right, for example, belongs to measure 3, despite all symbols but the anchor are sitting in the previous measure. This is because a segment always belongs to where its anchor is.


While in orchestral music chords are most often the result of multiple instruments playing together, in many popular music styles, chords are played on a single instrument, e.g. a piano, organ, guitar. The latter requires them to be notated as such in a single figure. We got two symbol types for chords:

  • Green (vertical scale, starting from chord root)
  • Purple (arpeggio, starting from lowest note in the current inversion of the chord).

If you want a figure to play narrow-voiced chords, the purple [a] symbols are superior to the green [c] chord symbols.


Synfire uses the purple chord symbols instead of the green, if you tick the Abstraction setting. It's named abstraction, because instead of attempting to faithfully recreate the original performance, it will replace the chord by its abstract model (pitch range, inversions) as provided by the harmonic context. In most cases this sounds more natural. 


Tip: You can convert any existing figure into one that uses abstraction with Transform >> Abstract Segments.

Full Chords

You can tag any vertical segment with two or more symbols to play the Full Chord as provided by the harmonic context. That is, even if the segment has only three symbols, it will play all the five notes of Am9, for example. You can use this to ensure the full harmony of your chord progressions is rendered by the figure.


Symbols before of after the anchor are considered relative to it. By setting the anchor, you decide which symbol is the most important. Because the anchor is rendered first, it falls closest to the symbol's notated position (e.g. a particular step of the scale), while the other symbols are subject to voice leading and other constraints, which tends to map them off their notated position.


Where the anchor should be placed:

  • If you want to make a melodic segment finish on a particular note, put the anchor at the end. Same goes for the other direction, respectively.
  • If the melody goes through an emphasized note, put the anchor in the middle at that note.
  • For vertical clusters and chords, the anchor is less relevant. Putting it on line zero is a good default. Experiment with it.
  • For the purple chord symbols [a], the anchor is least relevant, because all chord tones are almost always available and not subject to voice leading that much.
  • The anchor position also has benefits for the musical structure: An anchor at the end or beginning of a segment allows for the segment to reach beyond the bounds of the container, thus overlapping into the previous or next container.

Tip: Be aware that if you place an anchor exactly at the end of a figure vector (where the next loop begins), it will not be included with the loop.


Bass symbols are deliberately displayed bigger than others, because they often lie behind other symbols in the background. As is the nature of a bass line, bass symbols will often be at, or close to, the zero line, which denotes the bass note of the harmonic context.

In order to preserve melodic movement in the bass, you can group bass symbols in short segments. Choose the anchor at rhythmically accentuated positions, as is shown in the example below:


Too Many Bass Symbols?

of_bassinvasion.pngIf a polyphonic take converts to too many bass symbols, that is, most of the chords and melodies turn into bass symbols, this indicates the take is way off the playing range of the current instrument. Usually Synfire handles this automatically (the Transpose switch), but sometimes this isn't enough. 

  1. First verify if the playing ranges of the instrument are appropriate. If necessary, adjust them to roughly match the take's overall performance. Then try the figure recognition again.
  2. If that doesn't help, transpose the take parameter (or parts thereof) by an octave (click into take, deselect current selection and use the up arrow key to transpose 12 times). 
  3. Still no luck? Go and pre-tag symbol types manually in the take, as explained above.



By default, most figure recognition presets use accidentals to preserve as much detail as possible. Since the harmony assumed during figure recognition is only a guess (especially the scales), the resulting accidentals most often do not make sense in a different harmonic context.

We turned accidentals on by default in order to ensure a recorded figure best recreates the original performance in the same context. If the figure is used in a different context however, the accidentals are seemingly random and unmotivated.

In other words: Accidentals are not portable across a wide range of harmony.


Voice leading ensures that even the oddest accidentals play nicely (unless you disable voice leading and allow chromatic alterations). Although they don't do any harm, you may still want to disable or remove them from your figure to make it look simpler, easier to read and more portable across different harmony.

Tip: Accidentals are most useful for fixing or altering a melody in a fixed harmonic context (i.e. an almost finished song), where voice leading can be turned off for individual segments.


Average: 5 (5 votes)

Scholarly Lite is a free theme, contributed to the Drupal Community by More than Themes.