AVX — Acoustic Signal Vocabulary for Timed Text

Version 1.0 Specification · Published May 2026

Namespace URI https://cognispacehq.com/auravoxa/caption/1.0
Version 1.0
Published By Cognispace LLC
Section 1

Abstract

AVX is a metadata vocabulary for annotating timed text with acoustic speech signal observations. It is designed as a foreign namespace extension for TTML2 (W3C Recommendation, 2018) and as a JSON sidecar format for WebVTT. It is backward-compatible with all conformant TTML2 parsers. It is versioned. It is published and maintained by Cognispace LLC. This document specifies version 1.0 of the AVX vocabulary.

Section 2

Status of This Document

This document specifies AVX version 1.0, published May 2026. Cognispace LLC is the maintaining authority for this specification. This version supersedes no prior version. Future versions will be published at incrementing version paths under the same base URI.

This specification is published under the AVX Specification License (see Section 10). Implementation of this specification is permitted under that license. The AVX-Compliant certification mark is a separate trademark of Cognispace LLC and is not granted by this license.

Section 3

Namespace Declaration

Display the canonical namespace URI prominently:

xmlns:avx="https://cognispacehq.com/auravoxa/caption/1.0"

This URI is the normative identifier for the AVX 1.0 namespace. It is not required to resolve to a web page for the namespace to be valid in XML processing — however, this specification is published at this URI as a convenience to implementers.

The avx: prefix is the recommended prefix for this namespace. Implementations may use any prefix; the namespace URI is the authoritative identifier.

Section 4

Conformance

AVX Document

A conformant AVX Document is a TTML2 document that uses the AVX namespace and whose avx: attributes conform to the vocabulary defined in this specification. A conformant AVX Document must include the governance envelope attributes defined in Section 5.

AVX Renderer

A conformant AVX Renderer is software that reads AVX attributes and produces enriched output according to those attributes. A renderer that does not recognize avx: namespace attributes must ignore them silently and render the baseline tts: content normally. This requirement is not advisory — silent graceful degradation is a conformance requirement of this specification.

Section 5

Governance Envelope

These attributes appear on the root <tt> element. They are required in all conformant AVX Documents.

Attribute Type Required Description
avx:version string Required AVX vocabulary version. Current value: "1.0"
avx:governance string Required Governing specification. Value: "AVX-1.0" — refers to this document
avx:claim_boundary string Required Observation boundary declaration. The only permitted value in version 1.0 is "observational"
avx:context_type enum Required Content context. Values: live_unscripted, scripted_fiction, professional_review, educational

On avx:claim_boundary

The value "observational" is the only permitted value in this version of the specification. It declares that all signal values in the document are observations of acoustic properties of speech — measurements of how words were spoken — and are not claims about mental state, emotional state, psychological condition, deception, intent, or any other construct that cannot be directly measured from acoustic signal data alone.

Section 6

Signal Vocabulary

6.1 Utterance-Level Attributes

These attributes appear on TTML <p> elements. They describe properties of the utterance as a whole.

Attribute Type Values Acoustic Basis Description
avx:arc enum ASCENDING, SUSTAINED, RESOLVED, TAPERING, CARRIED, DROPPED Trajectory of acoustic energy and pitch movement across the utterance Direction of acoustic energy across the utterance. ASCENDING: energy builds toward the end. SUSTAINED: energy remains flat throughout. RESOLVED: energy rises then recovers before the end. TAPERING: energy descends in a controlled manner. CARRIED: energy does not resolve and continues into the next utterance. DROPPED: energy collapses abruptly mid-utterance.
avx:coherence float 0.0–1.0 Estimated alignment between acoustic delivery and lexical content Estimated alignment between how the utterance is spoken and what is said. Higher values indicate higher alignment. Lower values indicate the acoustic delivery diverges from the lexical register of the utterance. This is an observational estimate — not a claim about intent or truthfulness.
avx:effort enum low, medium, high Inferred pattern of acoustic effort in delivery, derived from prosodic emphasis distribution, voicing continuity, energy variance, and trajectory shape Inferred pattern of acoustic effort in this utterance. Derived from prosodic emphasis distribution, voicing continuity, energy variance, and trajectory shape. This is an acoustic inference — not a claim about the speaker's mental or emotional state.
avx:emphasis enum suppressed, distributed, concentrated, sparse Distribution pattern of prosodic emphasis across the words in the utterance Distribution pattern of prosodic emphasis across the words in this utterance. suppressed: emphasis is absent despite high voicing. distributed: emphasis is spread across multiple words. concentrated: emphasis is focused on one or two words. sparse: emphasis is present but minimal.

6.2 Span-Level Attributes

These attributes appear on TTML <span> elements. They describe properties of individual words or word groups.

Attribute Type Values Acoustic Basis Description
avx:eld enum low, medium, high, highest Relative acoustic energy for this word compared to surrounding words in the utterance Expressive load for this token relative to utterance baseline. Represents the distribution of expressive energy across the words in the utterance
avx:arousal float 0.0–1.0 Fundamental frequency and onset energy of this word Activation level. Derived from fundamental frequency and onset strength. Higher values indicate higher acoustic activation
avx:voiced float 0.0–1.0 Proportion of this word's duration in which the speaker's vocal folds were fully vibrating Proportion of voiced frames in this token. Higher values indicate more fully voiced delivery. Lower values indicate breathed, suppressed, or whispered delivery
avx:hedge boolean true, false Presence of hedge markers detected through lexical analysis of the utterance Whether this word was identified as a hedge marker — a word or phrase that introduces epistemic uncertainty, tentativeness, or qualification into the utterance.
Section 7

Complete Example

The following is a full TTML2+AVX document example showing all document-level, utterance-level, and span-level attributes in use.

<!-- tts: attributes = standard TTML styling layer (reads in any player) -->
<!-- avx: attributes = signal annotation layer (reads in AVX-aware renderers) -->
<?xml version="1.0" encoding="UTF-8"?>
<tt xml:lang="en"
    xmlns="http://www.w3.org/ns/ttml"
    xmlns:tts="http://www.w3.org/ns/ttml#styling"
    xmlns:avx="https://cognispacehq.com/auravoxa/caption/1.0"
    avx:version="1.0"
    avx:governance="AVX-1.0"
    avx:claim_boundary="observational"
    avx:context_type="live_unscripted">

  <head>
    <styling>
      <style xml:id="s1"
        tts:fontFamily="Arial"
        tts:fontSize="100%"
        tts:color="white"
        tts:backgroundColor="rgba(0,0,0,0.82)"
        tts:padding="4px 10px"/>
    </styling>
  </head>

  <body>
    <div>

      <!-- Utterance 1: tapering trajectory, high effort, suppressed emphasis -->
      <p xml:id="utt-01"
         begin="00:00:01.20" end="00:00:03.80"
         style="s1"
         avx:arc="TAPERING"
         avx:coherence="0.31"
         avx:effort="high"
         avx:emphasis="suppressed">

        <span tts:fontWeight="normal"
              avx:eld="low" avx:arousal="0.28" avx:voiced="0.91">I </span>

        <span tts:fontWeight="bold"
              avx:eld="high" avx:arousal="0.87" avx:voiced="0.22">know </span>

        <span tts:fontWeight="normal"
              avx:eld="low" avx:arousal="0.31" avx:voiced="0.88">you </span>

        <span tts:fontWeight="normal"
              avx:eld="medium" avx:arousal="0.40" avx:voiced="0.85">were </span>

        <span tts:fontWeight="bold"
              avx:eld="highest" avx:arousal="0.92" avx:voiced="0.19">there.</span>
      </p>

      <!-- Utterance 2: sustained trajectory, hedge present -->
      <p xml:id="utt-02"
         begin="00:00:04.50" end="00:00:08.10"
         style="s1"
         avx:arc="SUSTAINED"
         avx:coherence="0.38"
         avx:effort="high"
         avx:emphasis="suppressed">

        <span tts:fontWeight="normal"
              avx:eld="low" avx:arousal="0.44" avx:voiced="0.88" avx:hedge="false">That </span>

        <span tts:fontWeight="normal"
              avx:eld="low" avx:arousal="0.41" avx:voiced="0.89" avx:hedge="false">was </span>

        <span tts:fontWeight="normal" tts:fontStyle="italic"
              avx:eld="medium" avx:arousal="0.47" avx:voiced="0.87" avx:hedge="true">possibly </span>

        <span tts:fontWeight="normal"
              avx:eld="low" avx:arousal="0.39" avx:voiced="0.88" avx:hedge="false">the </span>

        <span tts:fontWeight="bold"
              avx:eld="highest" avx:arousal="0.91" avx:voiced="0.96" avx:hedge="false">most </span>

        <span tts:fontWeight="bold"
              avx:eld="highest" avx:arousal="0.93" avx:voiced="0.95" avx:hedge="false">significant </span>

        <span tts:fontWeight="normal"
              avx:eld="low" avx:arousal="0.38" avx:voiced="0.90" avx:hedge="false">thing.</span>
      </p>

      <!-- Utterance 3: ascending trajectory, fully voiced, building energy -->
      <p xml:id="utt-03"
         begin="00:00:09.00" end="00:00:12.40"
         style="s1"
         avx:arc="ASCENDING"
         avx:coherence="0.78"
         avx:effort="low"
         avx:emphasis="concentrated">

        <span tts:fontWeight="normal"
              avx:eld="low" avx:arousal="0.38" avx:voiced="0.88" avx:hedge="false">She </span>

        <span tts:fontWeight="normal"
              avx:eld="medium" avx:arousal="0.52" avx:voiced="0.90" avx:hedge="false">later </span>

        <span tts:fontWeight="normal"
              avx:eld="medium" avx:arousal="0.58" avx:voiced="0.93" avx:hedge="false">had </span>

        <span tts:fontWeight="normal"
              avx:eld="medium" avx:arousal="0.60" avx:voiced="0.92" avx:hedge="false">her </span>

        <span tts:fontWeight="bold"
              avx:eld="high" avx:arousal="0.82" avx:voiced="0.97" avx:hedge="false">entire </span>

        <span tts:fontWeight="bold"
              avx:eld="highest" avx:arousal="0.96" avx:voiced="0.98" avx:hedge="false">case </span>

        <span tts:fontWeight="bold"
              avx:eld="high" avx:arousal="0.88" avx:voiced="0.95" avx:hedge="false">dismissed.</span>
      </p>

    </div>
  </body>
</tt>
Section 8

Rendering Guidance (Non-Normative)

This section is non-normative. It describes how AVX attributes map to visual rendering decisions in reference implementations. Implementations are not required to follow these mappings exactly, but must conform to the governance constraints in Section 4 and Section 9.

Voice weight mapping (avx:eld)

avx:eld Recommended font-weight Recommended font-size
low40095%
medium500100%
high600108%
highest700116%

Color temperature mapping (avx:voiced)

voiced range Temperature Example color
< 0.35Cool#B8D8DC
0.35–0.70Neutral#E8DDD0
> 0.70Warm#F5C4A0

Italic mapping (avx:hedge)

avx:hedge Rendering
trueitalic
falseroman
Section 9

Out of Scope

  • AVX does not define how signal values are computed. The acoustic measurements that produce AVX attribute values are not specified by this vocabulary.
  • AVX does not define or constrain rendering behavior beyond the conformance requirement that unknown attributes be ignored silently.
  • AVX does not detect, identify, or classify emotional states.
  • AVX does not detect, identify, or assess mental states of any kind.
  • AVX does not assess deception, dishonesty, evasion, or intent to mislead.
  • AVX does not assess the truthfulness or credibility of speech.
  • AVX does not assess the psychological condition of speakers.
  • AVX does not provide clinical, therapeutic, diagnostic, or forensic outputs.
  • AVX attribute values are observations of acoustic properties. Cause is not determinable from acoustic evidence alone.
  • Renderers must not use AVX values to display indicators that imply judgment, scoring, ranking, or behavioral assessment of speakers.
Section 10

AVX Specification License

Copyright 2026 Cognispace LLC. All rights reserved.

Permission is hereby granted, free of charge, to any person or organization to read, implement, and distribute implementations conforming to this specification, subject to the following conditions:

  1. Any implementation claiming conformance to this specification must conform to all normative requirements in this document.
  2. This license does not grant permission to publish modified versions of this specification that claim to be AVX version 1.0. Modified versions must be clearly identified as derivative works and must not use the name "AVX" without a distinguishing qualifier.
  3. The AVX-Compliant certification mark is a trademark of Cognispace LLC. Use of the certification mark requires a separate trademark license. This specification license does not grant use of the certification mark.
  4. This license does not grant any rights under any Cognispace LLC patents.

THE SPECIFICATION IS PROVIDED "AS IS" WITHOUT WARRANTY OF ANY KIND.

Section 11

Version History

Version Date Summary
1.0 May 2026 Initial publication. Governance envelope, utterance-level attributes (avx:arc, avx:coherence, avx:effort, avx:emphasis), span-level attributes (avx:eld, avx:arousal, avx:voiced, avx:hedge), conformance classes, and AVX Specification License.
AVX v1.0 · Protocol Specification · Active Release · © Cognispace, LLC