LabsAI · Research Brief · Model Classes
Speech-to-Action, Speech-to-Orchestration, and Speech-to-Visualization.
August 16, 2026 · Updated September 2026
Speech-to-text made machines hear. Text-to-speech made them speak. Neither made them do, coordinate, or manifest. STA Models — Speech-to-Action —, STO Models — Speech-to-Orchestration —, and STV Models — Speech-to-Visualization — are three classes a spoken intelligent system needs the moment it stops merely talking and begins operating inside a live environment. One converts spoken intent into action. One converts spoken intent into the coordination of the intelligences doing the work. And one converts spoken intent into the visual manifestation of information, thought, state and interaction.
An STA Model converts spoken intent, conversation, application state and the vocabulary of available actions into action on a live surface — co-timed with the voice that explains it.
The competences that define the class are the ones a visitor actually feels: a referent resolving to the right on-screen object — “pause it,” “the second one,” “no — the other one” — a clarifying question on a genuinely ambiguous reference instead of a guess; and a report bound to a verified outcome — done meaning checked, not assumed.
Actras are the Labs family implementing STA. In generative terms, an Actra is a GAT — Generative Action Transducer.
STA predicts what happens.
An STO Model converts spoken intent into orchestration: which model or tool handles the work, what runs in parallel, what runs sequentially, when to retrieve, when to escalate, how to recover, how to verify and when to terminate — all of it steered by live speech at conversational tempo while a human holds the floor.
Its output is orchestration policy, and the policy can run as an AI Orch: the executable coordination object with participants, dependencies, state and a lifecycle.
Octras are the Labs family implementing STO. In generative terms, an Octra is a GOT — Generative Orchestration Transducer.
STA decides what the surface does. STO decides which intelligence does it — and how.
An STV Model converts spoken intent, conversation, information, application state, visual state and representational context into visualization on a live surface — capable of appearing and transforming while the conversation itself continues.
Its native output is not simply an image. It is visual representation. A spoken request may become a chart, diagram, interface, table, graph, simulation, generated scene, spatial composition, media surface, live webpage, interactive object or an entirely generated visual environment.
The defining problem is not simply: What should the system say? It is: What should this become visually?
An STV Model must determine what information deserves visual representation, which representation best carries the user’s intent, how that representation should be composed, when it should appear, how it relates to the spoken explanation and how it should transform as the interaction continues.
A user can say: “Visualize what you’re explaining.” “Turn that into a timeline.” “Show me the relationship.” “Make that a graph instead.” “Put those companies on a map.” “Build an interface where I can manipulate this.” “Show me what happens if this doubles.”
Those are not merely requests for image generation. They are Speech‑to‑Visualization commands against a living visual state.
Vistras are the Labs family implementing STV. In generative terms, a Vistra is a GVT — Generative Visualization Transducer.
STV decides what becomes visible — and how thought manifests on the surface.
The classes operate independently, but their greater value emerges when they operate together.
STA — Speech-to-Action
Speech → Action
STO — Speech-to-Orchestration
Speech → Orchestration
STV — Speech-to-Visualization
Speech → Visualization
A single spoken instruction can invoke all three. A visitor might say: “Find NVIDIA’s revenue over the last ten years, compare it with the growth of AI infrastructure, show me the inflection points, and let me manipulate the result.”
STO can determine which models, retrieval systems, data sources and tools should perform the work. STA can execute the retrievals, operate the surface and respond to subsequent commands. STV can determine that the result should become an interactive temporal visualization, construct that representation and transform it as the conversation evolves.
One sentence. Three prediction classes.
Orchestration. Action. Visualization.
Throughout the platform, this work arrives spoken. A visitor’s sentence can land as a:
Speech‑to‑Action Voice Command — the unit of spoken work an STA Model interprets and executes against the surface.
Speech‑to‑Orchestration Voice Command — the unit of spoken coordination an STO Model interprets and runs as policy.
Speech‑to‑Visualization Voice Command — the unit of spoken representation an STV Model interprets and manifests against the visual environment.
One sentence may contain several of each. The classes determine where each instruction belongs.
This means speech is no longer limited to producing another linguistic response. Speech can directly produce: action, orchestration and visualization.
STV introduces another state into the spoken intelligence loop. An intelligent system already needs to understand conversation state. STA requires knowledge of execution and interface state. STO requires knowledge of orchestration state. STV requires knowledge of visual state.
The system must understand not only what has been said and what has been done, but what currently exists visually.
That makes expressions such as: “Move that over there.” “Expand the second one.” “Keep this chart.” “Replace that table.” “Show those geographically instead.” “Make this three-dimensional.” “Now animate what happens next.” meaningful operations against a persistent visual environment.
Visualization becomes part of the conversational state rather than disposable output attached to a message.
Generative UI is one component of STV. Image generation can be another. Charts, diagrams, simulations, webpages, media and spatial environments can all participate. But STV is broader than any individual rendering technology. The model class is defined by the competence to infer: what should become visible.
That distinction matters. A conventional system can receive a command to “make a bar chart” and invoke a chart renderer. An STV Model should increasingly be capable of receiving: “Help me understand this.” and determining that a particular visual representation is the appropriate answer.
The representation itself becomes part of the reasoning problem. Visualization is therefore not merely presentation after intelligence.
Visualization becomes an output of intelligence.
The classes are stated in full in Larger Than Language; the voice loop that runs them is specified in Knowing When to Speak; and the existing action and orchestration classes are scored publicly on the Interactive Intelligence Bench — the Actra Score over the action-side tracks and the Octra Score over floor discipline, with a truthfulness penalty that scores a false “done” below an honest failure.
The first-generation STA surface and STO layer are live in LabsAI Studio. LabsAI Studio’s existing ability to compose charts, tables, cards, media and live surfaces during conversation provides an early production substrate for STV.
STV and the Vistra family formalize visualization as its own native prediction class. A dedicated Vistra evaluation framework should measure visualization competence independently rather than retroactively assigning STV performance from Actra or Octra scores.
STA and STO are also the first capabilities planned for a specialised domain. In FeMa — Frontier Editorial Media Architecture, now in development, a spoken editorial instruction such as “turn this conversation into a brief for the board” is designed to resolve as STO over research and drafting, and as STA over the editorial tools that carry it out. That domain is Frontier Journal, the media superapp of Labs Media Group and a Labs Centerpiece beside LabsAI Studio, where FrontierAI, LabsAI’s sister intelligence on LABA, is in development.
The families implementing the three classes are presented in the companion announcement:
Read Introducing Actras, Octras & Vistras →The distinction can be reduced to three questions.
STA: What should happen?
STO: Who or what should make it happen, and how should they coordinate?
STV: What should become visible?
And to three model families:
Actras act. Octras coordinate. Vistras manifest.
Together, they move spoken intelligence beyond the assumption that every intelligent operation must terminate in language. A machine can hear an idea and act. It can hear an objective and orchestrate. It can hear a thought and make it visible.
That is the larger transition:
Speech → Action
Speech → Orchestration
Speech → Visualization
Contributing authors: Maya E. Davis · Duránd F. Davis Jr.
This research is published while the questions are still open, and the systems it describes are live. We would love for you to join us — and please share your thoughts at research@labsintelligence.ai.
Please cite this work as:
Davis, Maya E., and Davis, Duránd F., Jr., “Introducing STA, STO & STV.” LabsAI Research Briefs, Labsintelligence — lab1 of Labs Companies, Inc., August 2026; expanded September 2026.
Or use the BibTeX citation:
@article{labsintelligence2026introducingstastostv,
author = {Davis, Maya E. and Davis, Duránd F., Jr.},
title = {Introducing STA, STO \& STV},
journal = {LabsAI Research Briefs},
publisher = {Labsintelligence, lab1 of Labs Companies, Inc.},
note = {LabsAI and its sister FrontierAI are built by Labsintelligence on LABA, LabsAI for LabsAI Studio and FrontierAI for Frontier Journal},
year = {2026},
month = {august},
note = {Expanded September 2026},
url = {https://research.labsintelligence.ai/labsai/introducing-sta-sto/},
}