Catlx Voice Pipeline

CATLX — Voice-First Pipeline

This skill owns the voice interface, which is the primary interaction model of CATLX — not a convenience layer. Every capability is accessible by voice; the keyboard palette and GUI are derivations.

Canonical detail: ../knowledge/references/voice-pipeline.md (+ detailed lifecycle in ../knowledge/references/runtime-lifecycle.md §18.2). Load on demand.

Purpose

Convert a spoken command, end-to-end, into a completed action: wake word → capture → transcribe → enrich with context → intent → disambiguate → plan → execute → speak → remember.

When to activate

• User asks how CATLX hears/understands/speaks, or about STT/NLU/TTS/wake word.
• Configuring a custom wake word (Wake Word Studio).
• Debugging a voice command that failed misinterpretation or disambiguation.
• Designing a multi-utterance conversation flow.

What this skill handles

1. The 10-stage pipeline (each stage an isolated module on typed message queues): Wake Word → Audio Capture → STT Transcription → Context Injection → NLU/Intent → Clarification → Planning → Execution → Feedback → Memory Write. (Stage table: ../knowledge/references/voice-pipeline.md §4.2.)
2. Wake word configuration — default Hey CATLX; custom via Wake Word Studio with 10–30 samples, compiled to a local ONNX model, fully on-device.
3. Continuous context — the Conversational Context Window (last N turns by tier, active app, last file, in-progress workflow). Injected into every NLU call, enabling pronoun resolution and implicit references.
4. Real-time responses — T1+: streamed TTS in parallel with workflow execution; T0: buffered TTS at completion to avoid CPU contention.
5. Lifecycle — detailed step sequence in ../examples/voice-command-lifecycle.md (wake word → memory write).

Requirements / constraints

• Voice-first mandate: other interfaces are convenience alternatives derived from the voice pipeline.
• Tier-scaled STT/TTS from the CapabilityMap (stt_engine, tts_engine, gui_mode); see catlx-hardware-adaptation.
• Least privilege (R5): the NLU step sends the context-enriched transcript through the PAL; the plan is validated by the Permission Router before execution.

Canonical knowledge it reads

../knowledge/references/voice-pipeline.md · ../knowledge/references/runtime-lifecycle.md · ../knowledge/references/hardware-adaptation.md.

Delegation

• AI provider for STT/NLU/TTS → delegate to catlx-ai-provider (skill({ name: "catlx-ai-provider" })). The PAL exposes transcribe(), complete(), synthesize().
• Executing the ExecutionPlan / running the DAG → delegate to catlx-workflow-engine (skill({ name: "catlx-workflow-engine" })).
• Writing the EpisodicRecord / memory context injection → delegate to catlx-memory (skill({ name: "catlx-memory" })).
• Choosing STT/TTS engine + fallback → delegate to catlx-capability-routing (skill({ name: "catlx-capability-routing" })).
• Speaking feedback / HUD → delegate to catlx-desktop-control (HUD) / catlx-electron-shell as appropriate.

Edge cases & warnings

• Low confidence / ambiguity: if confidence < 0.75 or ambiguity flags set, ask a clarifying question and loop — do not guess.
• T0 CPU contention: do not stream TTS during execution; buffer it.
• No cloud dependency: the wake word and (T0) STT/TTS run locally.
• Secure contexts: when the pipeline might touch sensitive input, respect input-synthesis/secure-mode (catlx-security).

Component lifecycle policy (reuse → install → adapt → create)

NEVER create a new component as the default. Before building/creating anything (a sub-skill, dependency, reference, workflow, helper, adapter, or template), check, in order:

1. Reuse an existing local component (resolve aliases/equivalent capabilities first) — reuse, don't rebuild.
2. Use an already-registered component from the registry.
3. Install a suitable existing, trusted, supported component → validate → register → connect to the graph → use.
4. Adapt an existing compatible component via a small persistent adapter/wrapper instead of re-creating it.
5. Create only as last resort — then make it permanent immediately: stable id, canonical location, register, add to the capability index + dependency graph, add provenance, use, and allow future reuse.
6. Never reorganise/recreate already-generated components (no Skill X 2 / new / temp variants); extend the existing one. Never create a second competing knowledge source; connect back to the canonical knowledge/ layer. Promote any reusable artifact out of /tmp/scratch into the permanent ecosystem.

Full policy: ../knowledge/rules/component-lifecycle.md.

Source / provenance

• Source: PART IV §4.1–4.6 (voice as primary interface, pipeline, wake word config, continuous context, real-time responses, lifecycle). Detailed lifecycle mirrored in PART XVIII §18.2.
• Inferred: none beyond Windows adaptation of audio/tray interaction; pipeline stages preserved exactly.