This document outlines a proposed standards framework for configurable voice interactions on the web, extending W3C efforts such as the Web Speech API [[WEB-SPEECH-API]]. The framework aims to standardize controls for voice-driven user interfaces (e.g., virtual assistants, podcasts, voice-enabled web applications), addressing gaps in timing, state management, context, user preferences, task handling, and error recovery.

Six API components are proposed: Temporal Control, Interaction State Machine, Context Management, Configuration Schema, Task Queue, and Recovery Protocol. Each builds upon or complements existing W3C standards to provide a coherent, interoperable platform for voice-driven web experiences.

This version of the document includes definitions of novel types and interfaces proposed for standardization. These types do not exist in any current W3C or WHATWG specification and represent new contributions requiring community review. A companion version of this document exists without these new type definitions for comparison purposes.

These notes are part of a Voice AI white paper by Paola Di Maio. They were shared for discussion with Patric and Francois of the W3C WebDX CG (via email, 29 January 2026) to explore possible usefulness and relevance to the WebDX Community Group.

This is a discussion draft and does not represent consensus of the W3C AI Knowledge Representation Community Group or any other W3C body. Feedback is welcome.

As well as sections marked as non-normative, all authoring guidelines, diagrams, examples, and notes in this specification are non-normative. Everything else in this specification is normative.

Proposed New Types New

This specification introduces the following novel types that do not currently exist in any published W3C, WHATWG, or other web standards document. These types are proposed as new additions to the web platform to support configurable voice interactions. Each requires community review, discussion, and formal standardization.

Temporal Control API

The Temporal Control API provides precise user control over spoken content, mimicking media player controls but applied to voice synthesis and AI-generated audio output. This enables users to perform actions such as skipping filler in AI responses or replaying important information.

This API builds on the existing Media Session API [[MEDIA-SESSION]] and extends it for voice-specific use cases.

[Exposed=Window]
interface VoicePlaybackController {
  undefined pause();
  undefined resume();
  undefined rewind(double seconds);
  undefined fastForward(double seconds);
  undefined setSpeed(double rate);
  double getCurrentTime();
};
        

Interaction State Machine

The Interaction State Machine defines the application states during voice exchanges, with user-triggered transitions. This ensures predictable UI feedback (e.g., visual indicators) and prevents confusion in real-time voice applications.

This API partially builds on the SpeechRecognition interface in [[WEB-SPEECH-API]].

enum VoiceInteractionState {
  "listening",
  "thinking",
  "responding",
  "processing"
};
        

[Exposed=Window]
interface VoiceInteractionStateMachine : EventTarget {
  readonly attribute VoiceInteractionState currentState;
  undefined transitionTo(VoiceInteractionState newState);
  attribute EventHandler onstatechange;
};
        

Context Management API

The Context Management API handles conversation continuity across sessions or interruptions. This is vital for complex tasks such as booking travel, multi-step workflows, or long-running voice sessions.

No existing W3C standard directly addresses this need; it represents a custom requirement for voice-driven web applications.

[Exposed=Window]
interface VoiceContext {
  undefined bookmark(DOMString label);
  undefined switch(VoiceContext newContext);
  undefined restore(DOMString bookmarkLabel);
  DOMString getSummary();
  Promise<undefined> persist();
};
        

Configuration Schema

The Configuration Schema defines a JSON-based structure for user and application preferences that customize voice interaction behavior. This builds upon the Web App Manifest [[WEB-APP-MANIFEST]] pattern of declarative JSON configuration.

{
  "pauseTimeout": 3000,
  "interruptionStyle": "patient",
  "thinkingMode": "explicit",
  "speedPreference": 1.0,
  "backgroundProcessing": true
}
        

Task Queue API

The Task Queue API manages asynchronous operations during voice sessions. It enables non-blocking task execution (e.g., sending an email while the voice assistant continues speaking), with status checks and cancellation support.

This API builds on existing patterns from Web Workers [[WEB-WORKERS]] and the JavaScript Promises model.

[Exposed=Window]
interface VoiceTask {
  readonly attribute DOMString taskId;
  readonly attribute DOMString type;
  readonly attribute any payload;
};

[Exposed=Window]
interface VoiceTaskResult {
  readonly attribute DOMString taskId;
  readonly attribute VoiceTaskStatus status;
  readonly attribute any result;
  readonly attribute DOMString? error;
};

[Exposed=Window]
interface VoiceTaskQueue {
  DOMString enqueue(VoiceTask task);
  VoiceTaskStatus getStatus(DOMString taskId);
  undefined onComplete(DOMString taskId, VoiceTaskCallback callback);
  undefined cancel(DOMString taskId);
};

callback VoiceTaskCallback = undefined (VoiceTaskResult result);

enum VoiceTaskStatus {
  "queued",
  "running",
  "completed",
  "failed",
  "cancelled"
};
        

Recovery Protocol

The Recovery Protocol provides resilience for voice sessions, restoring state after crashes, disconnections, or other failures. It builds partially on Service Workers [[SERVICE-WORKERS]] patterns for offline and recovery capabilities.

[Exposed=Window]
interface VoiceHistoryEntry {
  readonly attribute DOMString entryId;
  readonly attribute DOMTimeStamp timestamp;
  readonly attribute DOMString role;
  readonly attribute DOMString content;
  readonly attribute DOMString? state;
};

[Exposed=Window]
interface VoiceSessionRecovery {
  DOMString summarize();
  Promise<undefined> saveState();
  Promise<undefined> restoreState();
  sequence<VoiceHistoryEntry> getHistory();
};
        

Security and Privacy Considerations

Implementations of this framework MUST address the following security and privacy concerns:

• Context persistence: Stored conversation data MUST be protected with appropriate access controls. Implementations SHOULD encrypt persisted data at rest.
• Cross-origin concerns: Context and state data MUST NOT be accessible across origins without explicit user consent.
• User consent: Applications MUST obtain user consent before persisting conversation context or session history.
• Data minimization: Implementations SHOULD collect and retain only the minimum data necessary for the stated purpose.