Speech production requires a transformation from phonological representations into coordinated vocal-tract movements. This paper presents a spiking neural sensorimotor model within the Neural Engineering Framework and Semantic Pointer Architecture. Each syllable occurrence has a neural oscillator whose evolving phase coordinates its gestures and initiates the next syllable. Phase-dependent commands recruit separate neural populations for vocalic, oral-constriction, velopharyngeal, and glottal actions (gestures). Neural ramp integrators shape the rise and fall of gesture activation. Together with articulatory targets, these trajectories determine the extent and time course of articulatory movements. Opponent neural populations provide control signals to a geometric articulatory model. Thirty reference recordings cover an input corpus of eight lexical words, an alternative realization of one word, a syllable sequence, and isolated syllables. Two additional paired simulation subcorpora of sixty runs each, covering two words and one syllable sequence, assess activation reliability across neural realizations. Adjusting gesture command amplitudes increased successful realizations from 57/60 to 60/60 under the specified criteria, with ramp rate factors held constant. The model demonstrates how neural oscillator and ramp dynamics generate articulatory timing under specified phasing rules. It complements a preceding mental-lexicon model through an ordered syllable interface. Continuous online execution across all components and connection to the acoustic extension remain further integration steps. Accessible code and synchronized visualization support inspection of the neural activation trajectories and resulting vocal-tract movements.
Figures & tables
Figure 1: Functional parts of the production pathway. Vertical labels and brackets group cognitive processing through ordered syllable readout, sensorimotor processing through opponent neural control, and the articulatory-acoustic model. The ordered syllable sequence forms the cognitive-to-sensorimotor interface; decoded parameter trajectories form the sensorimotor-to-articulatory interface. Gray arrows mark interfaces implemented through symbolic readout, external program selection, or numerical transfer, and the planned onward acoustic connection. The dark arrow from opponent outputs to geometry marks the direct implementation in the exporter. The diagram describes the functional components and their interfaces; details of the numerical execution are given in A.5 . Schematic generated programmatically with AI-assisted code.
Gesture position
Start
End
Single onset consonant
0.1500 ( 54.00∘ )
0.4317 ( 155.41∘ )
Two-consonant onset: first
0.1500 ( 54.00∘ )
0.3848 ( 138.51∘ )
Two-consonant onset: second
0.2439 ( 87.80∘ )
0.4787 ( 172.32∘ )
Stressed single vowel
0.2927 ( 105.37∘ )
0.7152 ( 257.49∘ )
Unstressed single vowel
0.2927 ( 105.37∘ )
0.6683 ( 240.59∘ )
Two-vowel nucleus: first
0.2927 ( 105.37∘ )
0.6213 ( 223.68∘ )
Table 1: Prescribed gesture activation windows in normalized syllable phase, with degrees in parentheses. Both representations are rounded independently from the full-precision source values. The intervals define the phase commands from which neural activation and movement subsequently develop.
Figure 2: Gesture activation windows for Schrank (SAMPA transcription: SraNk ). The oral, glottal, vocalic, and velopharyngeal actions occupy overlapping phase intervals. The dorsal actions for N and k are separate occurrences even though they share a constriction target. Bar height is for display only; amplitudes differ by gesture class. The dotted line marks the neural oscillator feedback cutoff. Schematic generated programmatically with AI-assisted code.
Gesture class
Original cg
Revised cg
vg
Oral consonantal
1.00
1.00
0.20
Vocalic
0.95
0.95
0.05
Velum opening
0.70
0.90
0.05
Glottal opening (syllable onset)
1.00
0.70
0.20
Glottal opening (syllable coda)
0.60 / 1.00
0.70
0.20
Table 2: Gesture command amplitude and ramp rate factor. Amplitudes cg use the dimensionless 0–1 reference scale defined in Section 3.7 . Two sets of gesture command amplitude values are used to generate the paired simulation subcorpora to investigate parameter-sensitive activation failures (Section 7.8 ). For syllable-coda glottal opening, the original amplitude is 0.60 in bap--gum--dip--bap and 1.00 in Schrank . The ramp rate factor is identical in both subcorpora.
Figure 3: Sequential initiation and gesture recruitment in the SPA-controlled four-syllable example. (A) The controller initiates the first neural oscillator and subsequently advances through the positions in response to next-syllable phase signals. The arrows indicate initiation order; neural oscillator activity can overlap in time. (B) For each active neural oscillator, retrieval of the syllable identity stored for its sequence position releases the appropriate gesture pathways, through which that oscillator’s phase commands drive the gesture ramps. The diagram shows the functional sequence without assigning a duration to utterance preparation. The startup of each neural oscillator provides its syllable preparation time; overlapping neural oscillator activity is shown in Figure 7 . Schematic generated programmatically with AI-assisted code.
Figure 4: Phase reference for a neural syllable oscillator. The schematic shows the oscillator’s two-dimensional state space, using the same convention as the visual simulation player : 0∘/360∘ lies on the negative x axis, and phase increases counterclockwise. The red arrow indicates the direction of the initiating input to the first coordinate x . The circle is a phase reference, not a recorded neural trajectory. Schematic generated programmatically with AI-assisted code.
Word / realization
Program
G
Neurons
Min. peak
Tasse
ta → se
6
12,400
0.88
Tasche
ta → Se
6
12,400
0.88
Kanne
ka → ne
6
12,400
0.60
Katze
ka → tse
8
14,800
0.88
Teller (short ER)
tE → lA
5
11,200
0.88
Teller (long ER)
tE → lEA
6
12,400
0.81
Table 3: Neuron counts and gesture activation peaks in the nine recorded word realizations. G is the number of separate gesture occurrences. The neuron count includes neural oscillator, next-syllable, pulse, and ramp populations; each program additionally uses 5,400 neurons in the opponent stage. The final column gives the smallest peak decoded gesture activity within the recording, rounded to two decimals. Program codes are explained in A.3 .
Figure 5: Recorded realization of isolated bap from the original reference recordings. (A) Phase of its single neural syllable oscillator; samples with amplitude below 0.3 are omitted, and phase wraps are broken. (B) The four separate decoded gesture activations, including distinct labial onset and coda gestures. (C) Selected signed opponent outputs before geometric gains; negative vocalic-height values specify the lowering direction. (D) Stored midsagittal frames at four sample times; the tongue is highlighted in orange. All panels retain the timestamps and filtering of the same recorded cascade (Section 5.2 ). Visualization of stored simulation outputs using AI-assisted plotting code.
Figure 6: Recorded realization of Kanne . (A) Phase calculated from each decoded neural oscillator state; samples with amplitude below 0.3 are omitted, and lines are broken at phase wraps. (B) Separate decoded gesture activations. (C) Selected signed outputs of the opponent populations before geometric gains. Negative velum values represent the opening direction. (D) Stored midsagittal frames at four sample times; the tongue is highlighted in orange. All panels derive from the same recorded cascade. The curves retain the timestamps and filtering specified in Section 5.2 . Visualization of stored simulation outputs using AI-assisted plotting code.
Figure 7: Occurrence-specific activity in the adapted SPA-controlled sequence. The time axis includes the prescribed sequence-loading interval ( A.2 ). Top: amplitudes of the four decoded neural oscillator states. Bottom: selected labial gesture occurrences, including the final action of dip and the initial action of the repeated bap . These curves come from separate neural ramp populations. The current adaptation retains the earlier controller while separating gesture occurrences and applying the current phasing rules. Visualization of stored simulation outputs using AI-assisted plotting code.
Program
Original cg
Revised cg
Kanne
20/20
20/20
Schrank
20/20
20/20
bap--gum--dip--bap
17/20
20/20
Total
57/60
60/60
Table 4: Successful complete sequences in the two paired simulation subcorpora of 60 runs each. All prescribed neural oscillators activated in every run in both subcorpora. All three original failures concerned the velum ramp in slot 2 of the SPA sequence. The same three seeds succeeded with the revised gesture command amplitudes.
Figure 8: Velum activation in slot 2 of bap--gum--dip--bap for the three originally failed neural seeds. Gray: original commands; green: revised commands, with the same neural seed and vocabulary in each pair. These are 10-ms-filtered traces retained at 1-ms spacing. The dotted line marks 0.2; the activation criterion additionally requires at least ten consecutive samples above this level. Both velum and glottal gesture command amplitudes were changed together. Visualization of stored simulation outputs using AI-assisted plotting code.
Figure 9: Decoded gesture activity in the second syllable of Tasse (left) and isolated se (right), using revised gesture command amplitudes and seed 11. Columns show actual simulation time, without aligning the separate syllable onsets. The zero line distinguishes negative excursions from positive activation. Negative samples remain visible but receive zero weight at the articulatory target interface. These two illustrative recordings are separate from the two paired simulation subcorpora totaling 120 runs. Visualization of stored simulation outputs using AI-assisted plotting code.
Question
Coupled planning oscillators
Spiking Neural Model
How is timing organized?
Relative phases settle through coupling, conditional on the specified graph, coupling functions, and parameters.
Prescribed windows of syllable phase recruit separate gesture ramps.
What linguistic patterns are explained?
Accounts for selected consonant–vowel and cluster patterns, and acquisition preferences under stated learning assumptions.
Realization of supplied syllable programs; emergence and acquisition of their coordination patterns are not yet modeled.
How explicit is neural implementation?
Dynamical planning states; the learning model includes abstract neural units and Hebbian attunement, without an explicit spiking circuit.
Timing and recruitment are implemented with spiking populations and specified synaptic transformations.
What remains biologically unidentified?
The specific neural populations and connections realizing these planning states and couplings.
The human neural populations and codes corresponding to the implemented states and connections.
Table 5: Different explanatory emphases of coupled planning oscillators, particularly the account of Nam et al. (2009) , and the present Spiking Neural Model. The comparison concerns the cited and current implementations, rather than limits on what either framework could implement.
Aspect
Bekolay (2016)
Present implementation
Trajectory construction
Multidimensional forcing function decoded from neural oscillator phase.
Phase windows issue amplitude commands; each gesture occurrence has a separate recurrent ramp population.
Start mechanism
Default inhibition; a start-gated error population resets neural oscillator state.
Input or controller drive initiates each occurrence neural oscillator; no equivalent target-error reset population.
Downstream output
Vector gesture score converted for VocalTractLab; neural articulatory expansion left for future work.
Ramps weight articulatory targets, followed by opponent neural populations and geometric movement generation.
Table 6: Architectural comparison with Bekolay’s production implementation (2016, Section 6.2). Neural oscillator phase supplies a temporal reference in both models; the downstream construction of gesture trajectories differs.
Aspect
Modeling advantage
Cost or limitation
Interpretable control
Onset, duration, gesture command amplitude, and ramp rate factor can be adjusted for each gesture occurrence.
These controls interact; gesture command amplitude also affects rise slope and finite-window peak activation.
Gesture state
Each population retains activation history and can continue decaying after its command ends.
Residual activity and transmission delays require assessment; the present response law restricts available curve shapes.
Rate manipulation
Command timing can change while gesture response time is retained or adjusted separately.
Shorter commands can reduce activation; changing phase speed alone does not uniformly compress the full trajectory.
Neural implementation
Recruitment and dynamics can be inspected and perturbed separately for each gesture.
Additional populations and pathways require reliable drive; a successful neural oscillator start does not ensure every ramp is recruited.
Table 7: Advantages and costs of separate gesture-ramp populations in the present architecture. Empirical benefits require evaluation beyond the current simulation corpus recorded at a fixed nominal rate.
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
Quantity
Value
Neuron model
LIF
Neural oscillator
2 dimensions; 2,500 neurons per occurrence
Neural oscillator intercept distribution
Uniform on [0.3,1]
Nominal frequency; recurrent synapse
1.5 Hz; 25 ms
Neural oscillator amplitude threshold
Neural oscillator norm at least 0.3
Gesture pulse / next-syllable population
1 dimension; 200 neurons
Appendix
Table A.1: Principal neural parameters for networks with preselected syllables. Intercepts specify where neurons begin responding along their preferred direction. Seeds fix sampled neural parameters for a given construction; vocabulary vectors require their own specification (Section 8.3 ).
Quantity
Value
Input pulse (preselected syllables)
0.15–0.20 s; unit amplitude
SPA sequence loading (prescribed)
Four 100-ms syllable-position inputs
SPA start command
0.40–0.45 s
Timing-network simulation duration
2.5 s
Integration step
1 ms
Neural oscillator, ramp, initiation probe filters
10 ms
Appendix
Table A.2: Input, simulation, and observation settings. Probe filters define recorded estimates. Filtered gesture recordings also supply the input to the subsequent opponent stage.