How VoiceWiz Works

Every dictation in VoiceWiz travels the same route: Context → Transcription → optional Prompt Builder → output. Understanding the stages helps you configure the app to fit your work.

The pipeline

hotkey pressed
  └─ recording (tap-to-talk or push-to-talk)
       └─ transcription (local model or configured cloud service)
            └─ prompt builder (optional — mode + context + transcript)
                 └─ paste into active window
                      └─ saved to history

Stage 1: Active Context

Before any audio is processed, VoiceWiz reads the active context for the current mode — configurable slots that shape everything downstream. In research mode, for example, a slot can carry your working file, a topic, and the sources you're citing.

Stage 2: Transcription

Recording produces audio that is converted and sent to your configured transcription backend — a local model or a cloud provider you've set up with your own API key. The result is a raw transcript: accurate, but spoken-language messy.

Stage 3: Prompt Builder (optional)

If the active mode uses refinement, the transcript is combined with the active context and passed to your configured model as a structured prompt. This is where rough speech becomes usable output. In Direct mode this stage is skipped entirely: you get the raw transcript as-is.

Modes

VoiceWiz ships with five modes, each with its own behavior and slots:

Contexts and slots

Context is what separates VoiceWiz from a plain dictation tool. Each mode has configurable slots in the Control Panel — persistent bundles of settings that a dictation picks up automatically, so you don't re-explain your setup every time.

Writing slots

• A working file, so dictations continue the document you're writing with awareness of what's already there.
• Style direction for the refined output.

Research slots

Research mode has the richest slot configuration:

• Working file — the thesis chapter or paper on disk; VoiceWiz reads the current section as continuity context.
• Current section — e.g. “Chapter 3: Methodology”, so dictations land in the right place.
• Topic — an optional hint that anchors source gating.
• Sources — DOIs, URLs, freeform references, or local files. DOIs are enriched automatically with bibliographic metadata.
• Citation style — APA, MLA, Chicago, IEEE, or Harvard.
• Template — literature review, methodology, findings analysis, abstract, and more.

Switch modes from the Control Panel at any time; each remembers its own configuration. Pipeline slot configuration is stored locally on your machine.

The Prompt Builder

The Prompt Builder takes the raw transcript, combines it with the active context, and produces text shaped for the task — a polished paragraph instead of a stream of speech. It uses:

• The transcript of what you said.
• The active context — mode-specific slots such as a working file, sources, or style direction.
• A template chosen for the task — literature review, methodology, email, and so on.

When to enable it

• Enable refinement in Writing, Research, Programming, or Communication mode when you want task-ready output.
• Keep Direct mode (no refinement) when you want words exactly as spoken — plain dictation, notes-to-self, or filling forms.

How context influences output

• In Writing mode, prior text from your working file gives the model continuity.
• In Research mode, gated sources and relevant excerpts from your indexed literature ground the output — with citation checking to keep references honest.
• In Communication mode, the context keeps tone and format consistent.