Skip to content

Glossary

This page explains OpenScreech terminology in plain language. For file-format recommendations, see Output Formats Explained.

Capture and Recognition

Term Meaning
ASR Automatic speech recognition: converting spoken audio into text.
Activation type How a hotkey controls dictation. Push and Hold records while held; Toggle uses one press to start and another to stop.
Dictation A short, interactive voice capture normally started by a hotkey and delivered according to a profile.
Diarisation Estimating who spoke when in multi-speaker audio. It labels speaker ranges; it does not perform the speech recognition itself.
Hotkey The assigned keyboard control that starts and stops dictation and selects its starting profile.
Live transcription Recognizing audio during recording. Ordinary recording writes transcript outputs incrementally; multitrack recording assembles final exports after Stop.
Multitrack recording Recording separate channels from one physical audio interface on a shared timeline, with individual and combined transcript outputs.
Track name A persistent label in your multitrack list, independent of the interface. New unnamed tracks use Track N; a person’s name is optional.
Parakeet The local speech-recognition model used by OpenScreech.
Quick-switch key A key that changes profile during an active dictation. It does not start a session.
Selected text Text captured once from the frontmost application when an eligible profile hotkey starts dictation. A processor prompt can use it through {{SELECTED_TEXT}}.
Speaker-labelled transcript A transcript variant that adds speaker labels to the existing recognized words.

Profiles and Processing

Term Meaning
Auto Tag Deterministic phrase matching that attaches labels to a finished capture without rewriting its text.
Final output The original text or processor-stage result selected to continue into tags, routing, paste, files, Notes, or an HTTP response.
Input source The text a processor stage receives: the original post-replacement transcript or an earlier stage's output.
LLM Large language model: a model that can clean, rewrite, summarize, extract, classify, or structure text.
LLM post-processing Optional transformation of recognized text after word replacement.
LLM post-processor A reusable library entry containing a provider, endpoint, model, optional system-prompt delivery, timeout, credential, OpenAI-compatible custom headers, and optional wrapping.
Screechback Profile-controlled local text-to-speech playback of the final output or one configured LLM-stage result.
Speech model A text-to-speech model that converts text into audio. In OpenScreech this is distinct from a speech-recognition model and from an LLM.
Text source The final profile output or configured LLM-stage result selected for Screechback playback.
TTS Text-to-speech: synthesizing spoken audio from text.
Processor Input The single Meeting Recording transcript representation sent to the selected LLM processor as its user message. Original Transcript is the default; saved-output alternatives are optional.
Processor graph The connected set of LLM stages in a profile. It can be a sequence, contain branches, or combine both.
Profile A reusable live-dictation behavior preset controlling processing, delivery, saving, overlay behavior, and other session choices.
Stage One use of a selected LLM post-processor inside a profile graph. A stage chooses its own input source and can feed later stages.
Word replacement Deterministic text substitution performed before optional LLM processing.

Delivery and Storage

Term Meaning
Auto paste Automatically inserts the finalized result into the active application.
Flat-file Saved Outputs Profile-controlled audio and transcript files written into configured output folders, separately from Notes.
History A memory-only list of up to ten finalized results. It can be accessed from the menu or recalled with Left and Right Arrow during dictation.
Notes Persistent, searchable capture bundles containing metadata and selected transcript, tag, and audio artifacts.
Notes root The top-level folder the Notes window reads and profiles write into.
Paste format The clipboard representation delivered to another app, such as plain text, rich text, HTML, or multiple formats.
Paste input format The meaning OpenScreech assigns to text before clipboard conversion: plain text, Markdown, or HTML.
Request payload The complete request body sent to an LLM endpoint. A saved payload can contain transcript and rendered prompt content.
Sidecar A supporting file stored beside a main transcript or recording, such as word-level timing data.

Recording and Captioning

Term Meaning
Chunk A bounded section used to process or organize a long recording. Chunk duration affects segmentation, not the source audio's fidelity.
Cue One timed block of subtitle text. Cue splitting controls how much text appears in each block.
JSONL JSON Lines: one JSON object per line. OpenScreech uses it for machine-readable word timing data.
SMPTE-TT A broadcast-oriented form of TTML used by some professional media workflows.
SRT A widely supported subtitle format containing numbered text cues and time ranges.
TTML An XML-based timed-text format used in some professional caption delivery systems.
TXT A plain-text transcript without subtitle timing structure.
WebVTT A timed-text format designed for web and browser-based media playback.
Word-level timing Start and end timing information for individual recognized words.

Libraries, Automation, and Services

Term Meaning
Endpoint The network address that receives an HTTP or model request.
Library A reusable named collection or configuration that multiple profiles or workflows can reference.
Ollama A local model runtime that OpenScreech can use for LLM post-processing.
OpenAI-compatible An API shape implemented by various hosted and self-hosted services. It describes compatibility, not necessarily the service operator.
HTTP Transcription Service OpenScreech's local helper that accepts audio through an OpenAI-compatible transcription endpoint, and the dedicated window used to configure and monitor it.
Trigger A configured phrase that can match processed dictation text.
Trigger routing Matching transcript phrases and running explicit HTTP or shell actions.
Split routing Dividing one transcript around multiple trigger phrases so its segments can be routed separately.

Workflow Surfaces

Term Meaning
OpenScreech Home The central quick-action dashboard and embedded workspace for capture, processing, settings, and library features. Primary features can move between Home and separate windows without creating a second copy of their state.
App Settings The window for app-wide configuration such as microphone input, startup/session behavior, app presence, settings backup and transfer, local speech-model management, plus shortcuts to management windows.
Recording Settings The window for long-form recording, transcription outputs, diarisation, LLM processing, subtitle segmentation, storage, audio quality, and recording status.
Meeting Recording The long-form recording workflow with its own audio, transcription, caption, diarisation, and LLM-output preferences.
Transcribe File(s) The queue-based workflow for transcribing existing audio and video files with its own output preferences.

Quick Tips

  • “OpenAI-compatible” describes an API request shape, not who operates the service or where its data goes. Check the actual endpoint before sending sensitive text.
  • A profile governs live dictation and optional HTTP profile processing. Meeting Recording and Transcribe File(s) keep separate workflow preferences even when they share library post-processors.