The compositional idea
What if a recording could be reorganized not by cutting it at arbitrary points, but by listening to the family resemblances inside it?
A gesture may occur once as a short, bright attack and return later as a broader, darker shape. A phrase may be followed, many seconds later, by another that shares its movement but not its register, duration or intensity. These relationships are often hidden by the original chronology.
Gestural Lineage Composer analyses the recording, searches for related moments, and constructs a new order from them. It does not invent new notes or synthesize replacement material. It reconnects fragments of the original Sound, preserving their recorded timbre, attacks and internal movement while changing the larger temporal structure.
The central shift is from what happens next in the recording to what could follow this gesture musically.
Recurrence as memory
Hear a gesture return in altered circumstances. Similarity creates recognition; the differences between appearances give that recognition a sense of distance, memory or transformation.
Development as form
Connect locally related phrases into a trajectory that changes across time. The result can suggest intensification, thinning, darkening, brightening, or a gradual movement between sonic states.
Instead of asking whether two moments are identical, the composer asks: in what way are they related, and what new form emerges when that relationship becomes audible?
Two scales of musical thought
The analysis unit determines what the script treats as a compositional building block. The same Sound can yield different forms depending on whether it is heard in local gestures or larger phrases.
Short, fixed-duration excerpts reveal local contours: attacks, swells, changes of brightness, and recurring shapes within a continuous texture.
The recording is segmented at significant pauses and acoustic changes. Longer units can retain internal rhythm, articulation and the shape of a complete musical thought.
Use an existing Praat TextGrid to decide what counts as a phrase. Your own analysis becomes the starting point for computational recomposition.
Automatic boundaries are proposals, not definitive musical truths. If the material has ambiguous phrasing, compare the automatic result with a TextGrid of your own. When automatic segmentation cannot find at least two usable phrases, the script can switch to gesture-window analysis and reports that fallback.
From similarity to evolution
Finding repeated material is only the beginning. The script also asks whether related gestures can form a directional path: a sequence in which each step is connected to the next, but the overall character of the sound may change.
Shape: what moves in a similar way?
Two phrases may share a rising-and-falling intensity contour, or a similar pattern of spectral movement, even if one is higher, brighter or louder than the other. Shape emphasizes those related trajectories.
Absolute: how similar is the sound itself?
Absolute retains differences in the measured acoustic level or register. It tends to favor nearer acoustic states rather than similarity of contour alone.
The script listens through three measurable dimensions: spectral centroid (a rough indicator of brightness), loudness (level and dynamic contour), and spectral flatness (the balance between tonal and noise-like energy). The default comparison uses all three, but you can focus on one.
In phrase mode, contours of different durations can be compared on a shared analysis grid. Their actual audio durations are preserved, and phrase duration and internal silence also influence the relationship measure.
Four ways to compose with a lineage
Choose a mode according to the kind of form you want to explore. All four work by selecting and reordering excerpts of your original recording, not by generating substitute sounds.
Lineage reconstruction
Follow one sequence of related gestures or phrases. Best for hearing a single relationship become a musical process: recurrence, gradual change, or a continuous passage between connected states.
Compositional question: how does one idea acquire a history?
Evolutionary montage
Interleave multiple discovered paths, generation by generation. Rather than organizing one family, the result juxtaposes parallel developments and may produce dialogue between them.
Compositional question: how can several related processes coexist?
Branching lineages
Begin from a lineage and explore alternative continuations. The resulting form suggests divergence: a gesture that could develop in more than one way.
Compositional question: what other future could this idea have?
Guided exploration
Specify an anchor time and ask the system to build a path through that moment. In gesture mode, a selected TextGrid can also supply the anchor from its first labelled interval.
Compositional question: where can a particular sound lead?
Creative starting points
Presets suggest different ways of listening to and shaping the same material. They are starting points, not fixed musical styles. Choose one, hear its result, and then compare it with another interpretation of the recording.
Hidden Recurrences
Seeks closely related material and favors continuity. Useful when the source contains repetitions that are difficult to hear as a single family in their original context.
Gradual Evolution
Explores a path of sustained acoustic change. Try it when you want the source to move progressively from one sound character toward another.
Distant Relatives
Uses the more selective chain analysis to seek related gestures far apart in the recording. A useful way to examine recurrence across a long temporal span.
Branching Memory
Looks for alternative developments around a related origin, making room for repetition, deviation and competing continuations.
Sparse Lineage
Allows pauses between reconstructed units. Silence becomes part of the new articulation rather than an incidental gap in the source.
Custom
Start from your own choices of analysis unit, direction, mode and comparison features. Most useful after discovering what a particular recording offers.
Presets configure the analytical and compositional settings but do not make every source sound alike. Their results depend on the particular relationships in the selected recording. Distant Relatives uses Strict Chains with gesture windows. Phrase analysis always uses Evolution Path, including when that preset is selected.
Your first exploration
Begin with a recording that contains some recurring identity: repeated instrumental gestures, evolving electronic textures, voice fragments, percussion figures, or related phrases separated by pauses.
- Open a Sound in Praat. Select exactly one Sound object. The tool works from the recording already present in the Objects window.
- Choose an analysis unit. For material with clear musical statements, try Phrases (automatic). For continuous sound evolution or smaller details, try Gesture windows.
- Choose a compositional direction. Start with Auto (strongest steady change) or use Continuity only to hear related material without imposing a direction of change.
- Choose a mode or preset. Start with Lineage reconstruction. Use Evolutionary montage or Branching lineages when you are ready to explore more complex forms.
- Run, listen, and read the picture. Praat creates a reconstructed Sound when a usable path is found, and the optional visualization shows the acoustic material, its relationships, and the selected sequence.
- Compare interpretations. Run the same recording as gestures and as phrases. Keep the setting that reveals the most compelling musical relationship, not necessarily the one with the lowest statistical p-value.
Choosing your musical direction
The main dialog keeps the creative choices near the surface. Advanced settings offer more detailed control when you want to shape the behavior of the search and the assembly.
| Main setting | What it means for the music |
|---|---|
| Analysis unit | Gesture windows, automatically detected phrases, or phrases defined in a TextGrid. This sets the scale of the form. |
| Match on | Use all three acoustic dimensions, or focus on brightness, loudness, or tonal/noise-like character. |
| Evolve toward | Choose a directed change, automatic selection of a steady change, or Continuity only for a path that is not tied to one acoustic destination. |
| Window s | Duration of local gesture windows. This affects gesture analysis, not variable-length phrase analysis. |
| Mode | One lineage, a montage of several lineages, branching alternatives, or a path guided by an anchor. |
| Anchor time s | Directs Guided exploration toward a moment you choose in the Sound. |
| Draw visualization / Play result | Choose whether Praat draws its analytical picture and plays the reconstructed Sound. |
Available directions
Advanced controls worth knowing
Phrase detail influences how readily automatic segmentation introduces boundaries. Min / Max phrase control the intended scale; zero requests adaptive durations. Phrase boundaries allows pauses plus acoustic change, pauses only, or acoustic change only. TextGrid tier identifies the interval tier in the selected TextGrid.
Links per unit and Link tolerance affect how many candidate relationships are available. A more permissive setting can create longer, less selective paths. Silence between and Silence scale shape the spacing of reconstructed gestures. Crossfade ms smooths joins, while Max output s places an upper limit on the planned duration.
Surrogates and Strict govern statistical evaluation, not musical quality. Leave Strict off for exploratory composition: a usable path can be worth hearing even when its original temporal order is not statistically exceptional.
Reading the visualization as a score of relationships
The Praat Picture is more than a diagnostic graph. It gives the composer a view of how recorded time has been converted into compositional time.
| Panel | What you see | How to use it |
|---|---|---|
| A · Acoustic trajectory | Source feature over time, highlighted source fragments and, in phrase mode, phrase boundaries. | Ask which parts of the original recording have become the building blocks of the new piece. |
| B · Relationships | In phrase mode, an arc-based map of related phrases, with path links distinguished; in gesture mode, the corresponding relation or profile view. | See whether a path moves between nearby occurrences or connects distant moments that were not adjacent in the source. |
| C · Evolution path / lineage | Selected units and their directional connections, with acoustic-change and relation statistics. | Read the path as a proposed formal trajectory, not as a guaranteed musical judgment. |
| D · Reconstruction timeline | The newly assembled output sequence and its source correspondences. | Examine the resulting articulation, pacing, and placement of returns or divergences. |
In phrase mode, the script can also create a Praat TextGrid of the analysed phrases, with selected path members marked. This lets you inspect boundaries closely, relate the computed path to your own segmentation, and continue working with the material in Praat.
Compositional approaches
The tool is most productive when its analysis is treated as a suggestion for listening, not a command to accept the first result. Several contrasting approaches are possible.
One idea in several forms
Use a source with repeated figures, such as struck or plucked sounds. Let phrase-level comparison reveal a family of related gestures, then listen to how their reordering alters the sense of continuation and closure.
Change without discontinuity
Choose a directed evolution, such as Brighter or More tonal, with Shape similarity. Listen for a trajectory that changes its acoustic character while retaining something recognizably related from phrase to phrase.
Composition through absence
Try Sparse Lineage or proportional silence. The pauses can separate related events into a reflective sequence, where each return is heard against the memory of the previous one.
From personal analysis to new form
Mark phrase boundaries in a TextGrid according to your own listening. Let the script propose relationships among these larger units. This combines the composer's structural hearing with computational discovery.
The point is not to prove that the algorithm has found the “correct” sequence. It is to discover a sequence worth listening to, questioning, and developing further.
Requirements and practical notes
Use: select one Sound object in Praat and run Gestural_Lineage_Composer.praat. It calls its companion Python script gestural_lineage_engine.py. The engine is searched first in the Praat AudioTools plugin's py folder, and otherwise in Praat's working directory. Keep the matching scripts together to avoid loading an incompatible engine.
Python packages: the Evolution Path and phrase modes need NumPy, SciPy and SoundFile. The optional Strict Chains engine additionally needs STUMPY.
pip install numpy scipy soundfile
# Only if you intend to use Strict Chains:
pip install stumpy
Sound reconstruction: the script extracts original source audio across all channels and joins the selected excerpts, with optional crossfades and silence. It does not automatically pitch-shift, time-stretch, or spectral-process the output.
| What you encounter | What it means or what to try |
|---|---|
| Automatic phrases produce a gesture-window result | Too few usable phrases were detected. The script reports the fallback. Try adjusting Phrase detail or use a TextGrid to set meaningful boundaries. |
| No relationship between phrases | The phrase graph contains no acceptable multi-phrase path. Try Links per unit or Link tolerance, or compare at gesture level instead. |
| The result sounds fragmented | Try larger phrase units, reduce unnecessary segmentation, or compare crossfade and silence settings. Phrase boundaries may need a closer listen. |
| A strong path is labelled Exploratory | The path is musically usable but its original temporal order was not unusually strong in the statistical comparison. With Strict off, you can still hear and use it. |
| A TextGrid is selected but not used as phrase boundaries | Choose Phrases (TextGrid) as the analysis unit and, in Advanced settings, verify the desired interval tier. Merely selecting a TextGrid is not enough. |
| Python engine cannot be found | Check that gestural_lineage_engine.py is in plugin_AudioTools/py or beside the script in the current working directory. |