Flute-like KlattGrid Serial Melody — User Guide

What this does

Flute-like KlattGrid Serial Melody combines two mechanisms: a twelve-tone pitch/rhythm generator and a synthetic flute-like voice. In Melody mode it creates a 12-note phrase; with Melody mode off it renders one 2.5-second note at the exact Seed frequency.

Twelve-tone pitch process

The fixed seed row is:

These are pitch classes relative to C. The row contains all 12 pitch classes exactly once, but it is not an all-interval row. The older guide's “Webern Op.24 style / all-interval” description was incorrect.

What Seed frequency means in Melody mode

Seed frequency is a pitch-class anchor, not necessarily the first sounding frequency. The script converts the entered frequency to the nearest equal-tempered MIDI note and uses that note's pitch class to transpose the seed row. Base octave then controls the register used for the first transformed row member.

For Prime and Inversion, the transformed row begins on the anchor pitch class. Retrograde and Retrograde-Inversion begin with the final member of the corresponding forward row, so their first sounding pitch class is not the seed pitch class.

Row forms

Register placement

The 12 pitch classes are converted to equal-tempered MIDI pitches. After the first pitch is placed from Base octave, each following pitch class is placed in the nearest octave to the preceding note; if the direct placement is more than six semitones away, the script shifts it by an octave. The final tessitura is constrained to MIDI 48–96 (C3–C7).

Interval-derived rhythm

The rhythm is serially related to the transformed pitch row, but it is not a direct proportional mapping from semitone size to duration. The script first creates five evenly spaced rhythmic values between Min IOI and Max IOI:

palette = [Min,
           Min + 1/4 range,
           Min + 2/4 range,
           Min + 3/4 range,
           Max] × Duration scale

The directed pitch-class interval to the next row member then chooses one of those five values:

For notes 1–11, the selected value is the true onset-to-onset interval (IOI). Note 12 uses the directed cyclic interval back to row member 1 to select the final duration of the phrase.

Flute-like KlattGrid voice

Each note receives its own KlattGrid. Pitch is placed in the KlattGrid pitch tier, with a small first-note approach to the target and sampled sinusoidal vibrato. Voicing, aspiration, breathiness, and flutter are controlled by their corresponding KlattGrid tiers.

Pitch-tracking resonance ladder

The oral resonances are intentionally tied to F0 rather than representing measured fixed flute resonances:

The script works at a fixed 44.1 kHz sample rate and reserves practical headroom at 0.45 × Fs. A single note, or any generated melody, is rejected if the 8×F0 resonance would exceed that limit.

After KlattGrid synthesis the main tone is Hann-band filtered from 40 Hz to an adaptive upper frequency: the larger of 8 kHz or approximately 1.12× the fifth resonance, capped at the practical headroom frequency.

Vibrato and note envelopes

Pitch vibrato is specified in semitones and converted multiplicatively to frequency. The amplitude vibrato is correctly applied as an additive dB deviation around Voicing level rather than multiplying the absolute dB value.

The first note has a short pitch approach, voicing/aspiration/breathiness attack, sustain, and release inside its KlattGrid. Later legato notes begin at their sustain levels immediately; their inter-note fade is supplied by the external overlap window. The final note is ended by the script's short phrase-edge protection fade.

First-note chiff

Chiff is added only when do_attack is true — the first note of a melody, or the standalone note in single-note mode. Later legato notes contain no new chiff.

The three layers are summed directly into the mono main tone by sample index. There is no temporary multichannel combine/downmix trick in the current version.

Random seed controls all stochastic breath/chiff synthesis. Zero uses an unpredictable state; a positive seed makes the stochastic components reproducible, after which Praat's unpredictable random initialization is restored.

Legato overlap and phrase timing

The crossfade is:

min(50 ms, 20% of the minimum scaled IOI)

For every note except the last, that overlap is added to the note's sounding duration. The next note is nevertheless shifted to its true serial onset, so the two sounds overlap for the crossfade interval.

The amplitude windows are complementary raised-cosine curves:

fade in  = 0.5 - 0.5*cos(pi*t/X)
fade out = 0.5 + 0.5*cos(pi*t/X)

They sum to 1 at corresponding positions. This is a smooth complementary amplitude crossfade; it should not be confused with the sin/cos equal-power panning law used in some other AudioTools processors.

Presets

Presets set the seed/register, IOI range, Duration scale, KlattGrid voice controls, and chiff balances. They do not change Melody mode, Row form, Normalize output, Draw visualization, or Play result.

All current presets use Duration scale = 1.0. They also set amplitude-vibrato depth, aspiration attack peak, and breathiness level; those values remain editable on the advanced page.

Controls

Compact launcher

Advanced voice/chiff page

The advanced page exposes Open phase, pitch-vibrato rate/depth, amplitude-vibrato depth in dB, voicing level, aspiration sustain and attack peak, breathiness level, three chiff peak values, and Random seed. It opens after the selected preset is loaded, so it fine-tunes that preset rather than replacing it with generic defaults.

Level, channels, and output

Each synthesized note is intentionally peak-calibrated to 0.72 before overlap. This is a per-note calibration step, not the final phrase normalization. Chiff A, B, and C are also individually peak-calibrated to their requested peak values before they are mixed with the main first-note tone.

After the phrase is assembled, a short outer linear fade of at most 8 ms protects the beginning and end. If Normalize output is enabled, the complete result then receives Scale peak: 0.92, i.e. target peak normalization. With normalization disabled, the post-overlap raw phrase level is retained.

Visualization and QC

The QC strip reports the KlattGrid mechanism, row form and 12/12 pitch-class uniqueness, phrase duration and overlap, vibrato, aspiration, seed state, and pre/post-normalization levels.

Further Reading

• Klatt, D. H. & Klatt, L. C. (1990). “Analysis, synthesis, and perception of voice quality variations among female and male talkers.” Journal of the Acoustical Society of America, 87, 820–857. DOI: 10.1121/1.398894. Relevant to the Klatt source/filter parameter family on which Praat's KlattGrid is based.
• Praat Manual — KlattGrid. Documents KlattGrid as a time-varying source/filter model with pitch, phonation/noise, and vocal-tract filter tiers.