e↗entertrainment
← All project stories
Creative tools / PROJECT STORY

Vocal Expression module (model1 / VXR core)

The Vocal Expression module is the subject-grounded core of model1: it takes the app's live reading of a person's own voice, turns it into a word-free 'support stance', and either voices that stance back, hands a meaning to another speaker in their own vocabulary, or routes it to assistance actions.

Research module2 MIN READDEVELOPMENT ARCHIVE · OCT 2026
Vocal Expression · 12.3-second silent interface recording of the real Sonic Dialogue Test Bed (model1), 18 October 2026. Built-in synthetic sample, Klatt voice path, waveform, spectrogram and event log. Recorded screenshots encoded as video; no captured audio and no live microphone input.
Open video ↗
Real native Sonic Dialogue Test Bed · synthetic sample and generated phrase, with support-intent lanes, waveform, mel/PCEN spectrogram and MIDI 2.0 per-note controls. Live mic off · 18 October 2026.
Real native Sonic Dialogue Test Bed · synthetic sample and generated phrase, with support-intent lanes, waveform, mel/PCEN spectrogram and MIDI 2.0 per-note controls. Live mic off · 18 October 2026. Open full size ↗
Real native translation log for the same synthetic sample. The app reports its intent, utterance and rendered-phrase stages; this image is interface evidence, not an audio listening test.
Real native translation log for the same synthetic sample. The app reports its intent, utterance and rendered-phrase stages; this image is interface evidence, not an audio listening test. Open full size ↗

Run the project’s nine WebAssembly fixture modules and inspect their JSON responses, text routing and display previews. Uses synthetic deterministic values; no microphone, native app capture or live analysis.

Try the WASM fixture explorer ↗
PROJECT REVIEW · 18 October 2026

Inside the project

Prepared with the project’s development agent from source and retained evidence. This is a project-owner assessment, not an independent customer review. Dated checks were not rerun for this website.

The Vocal Expression module is the subject-grounded core of model1: it takes the app's live reading of a person's own voice, turns it into a word-free 'support stance', and either voices that stance back, hands a meaning to another speaker in their own vocabulary, or routes it to assistance actions. It is implemented and component-tested end to end in source; it has not been run live in this inspection, and its signals are assistive proxies, never clinical readings.

On 18 October 2026 the real native Sonic Dialogue Test Bed was captured using its built-in synthetic sample with the live microphone off. The screenshots show the generated phrase, support-intent lanes, waveform, spectrogram, MIDI 2.0 controls and translation log. A 12.3-second silent interface recording accompanies them. This demonstrates the native sample workflow visually; no audio listening acceptance, live voice interpretation or clinical result is established. A separate browser fixture explorer executes nine real WebAssembly modules with deterministic synthetic responses. It demonstrates the response contracts and display routing, not the full native analysis engine.

What stands out

    Current limits

      Features & evidence

      Subject-grounded event mirror

      component-tested

      Reads the live model's per-utterance signals/quality/tuning for the speaker; replayable from a logged event or synthetic sample.

      Recorded evidence

      VXRTunedAssistanceEvent (.sample + parsingLog init); mirrors VocalExpression.swift DerivedRegulationSignals/AssistanceTuning; swiftc/XCTest.

      Meaning map (7 support lanes)

      component-tested

      Deterministically turns the voice state into a word-free support stance; quality gates assurance.

      Recorded evidence

      VXRSupportIntent.from(event:); Tests/AssistantAppTests/VXRVocalResponseTests.swift.

      Codebook representation (8-dim → word-units)

      component-tested

      Recurring expressions quantise into reusable, named vocal units.

      Recorded evidence

      VXRExpressionFeatures (dim=8) + VXRExpressionCodebook k-means swiftc self-test.

      Recipient routing (shared lexicon translate)

      component-tested

      Voice a meaning in another speaker's vocabulary.

      Recorded evidence

      VXRLexiconTranslator.translate; test-bed translateLast; standalone 2-speaker verification.

      Action routing to assistance (node → tuning → execution)

      source

      Hold the mic on a node; its expression sets assistance tuning + vox: tags the execution layer consumes.

      Recorded evidence

      NodeVocalExpressionController:118 → AssistancePolicyMapper.map → node.vocalExpression; VocalShapedSupportBridge.swift:42. Runtime capture NOT verified.

      Review updates & corrections
      • No change to 17 Oct family maturity; this object adds the module-internal representation + routing detail your follow-up requested.
      • Re-verified in current source this session: VXRLexiconTranslator.translate (recipient routing); NodeVocalExpressionController → AssistancePolicyMapper + node.vocalExpression + vox: tags, and VocalShapedSupportBridge (action routing into the task execution layer).
      • Source hygiene note (not website copy): a commented-out duplicate 'VocalExpression 2.swift' exists alongside VocalExpression.swift — inert, but a known '… 2.swift' build hazard if un-commented.
      • On 18 October 2026 the real native Sonic Dialogue Test Bed was captured using its built-in synthetic sample with the live microphone off. The screenshots show the generated phrase, support-intent lanes, waveform, spectrogram, MIDI 2.0 controls and translation log. A 12.3-second silent interface recording accompanies them. This demonstrates the native sample workflow visually; no audio listening acceptance, live voice interpretation or clinical result is established. A separate browser fixture explorer executes nine real WebAssembly modules with deterministic synthetic responses. It demonstrates the response contracts and display routing, not the full native analysis engine.
      Planned captures · not yet available
      • Shared tab: seed a second speaker, Align, Translate-last → the same meaning voiced in the other speaker's vernacular.
      • Node-map: hold the mic on a node → capture → support popover with vox: tags and applied assistance tuning.

      Explore the connections.

      Browse all project stories ↗