> ## Documentation Index
> Fetch the complete documentation index at: https://docs.wittify.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Transcript Editor
> The per-file editor. Drag waves between speakers and to new times, edit words inline, watch the active word highlight while audio plays, and download the transcript or audio. Full undo / save / discard flow with an unsaved-changes guard.
The transcript editor is the deepest surface in Speech to Text. You reach it by clicking any row in the [File Library](/en/Systems/Speech-to-Text/File-Library) or any file in the **Your Files** sidebar. It shares the standard dashboard shell and uses a two-panel layout: transcript and timeline on the start edge, settings and tabs on the other side.
## Page layout
The page is split into two panels. On screens narrower than `md` they stack with the transcript on top and the settings panel underneath.
| Panel | What it holds |
| -------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Header bar** (top) | File name + filename, then a **Clock** chip showing the live audio length, a **Globe** chip showing the language, **Speakers** + **Subtitles** chips when enabled, the file **Status** pill, and on the end edge a **Transcript** download dropdown (TXT / Word / PDF) and an **Audio** download button. |
| **Editor toolbar** (between header and transcript) | **Merge**, **Split**, **Reassign** (dropdown of speakers), **Undo**, **Save changes**. The buttons enable based on what you have selected. |
| **Transcript** (left, scrollable) | The list of speaker turn rows. Click any row to select it. Click any word to edit it. The audio timeline sticks to the bottom of this panel. The **Save bar** slides in below the timeline whenever you have unsaved edits. |
| **Settings** (right) | **Transcribe Settings** at the top, then a tab switcher between **Summary** and **Intent**, then **Cue Properties** at the bottom. |
The breadcrumb at the top of the page reads **Speech to Text (Faseeh)** › *file name*. Renaming the file from the [File Library](/en/Systems/Speech-to-Text/File-Library) 3-dot menu updates this breadcrumb instantly.
## Default speakers
Two speakers ship by default: **Speaker 1** and **Speaker 2**. Each name is editable in place via the start-edge input on its **Timeline Row**. The name you type is what appears everywhere: the transcript bubble label, the side-panel **Summary** rows, the **Reassign** dropdown in the toolbar. Names render correctly in mixed Arabic and Latin scripts.
The transcript text itself can be in any language. The dashboard chrome (button labels, dialog copy, toast text) translates between English and Arabic, but the transcript content stays exactly as the audio came in — Arabic, English, a third language, or a mix.
## Header — Downloads
Two downloads in the file header:
A dropdown with three formats:
* **Plain text (.txt)** — UTF-8 with BOM so Excel can open Arabic correctly.
* **Word (.doc)** — opens directly in Microsoft Word.
* **PDF** — opens the formatted transcript in a new browser tab. Use **File → Print → Save as PDF** in your browser to save it.
For very long transcripts (over 5,000 messages) the export is truncated and a yellow notice appears at the bottom of the file.
A single button that downloads the original audio file. While the file fetches, the icon swaps to a spinner and the button is disabled. If the audio can't be reached you'll see a specific message ("Couldn't reach the audio file" for network issues, "Audio file unavailable (HTTP X)" for permission / not-found errors).
## Transcript pane
### Speaker turn rows
Each row is one speaker turn. The row carries:
* A **selection checkbox** at the start edge. Click anywhere on the row (except a word) to toggle selection.
* A **vertical accent line**: blue for **Speaker 1**, green for **Speaker 2**, red when the turn has an error.
* A **speaker label** showing the current speaker name (whatever you typed in the timeline), with the start time of that turn formatted `MM:SS Sec`.
* The transcript text as a list of clickable words separated by spaces.
A small fold-vertical separator icon appears between every two rows.
### Editable words
Every word is independently clickable. Click any word to flip it into a small inline input:
* The input grows or shrinks to fit the text you type.
* The background tints to a light brand color and the input has a thin brand border so you know it's active.
* Press **Enter** or click anywhere outside the input to commit the change. Press **Escape** to cancel the edit.
Editing a word makes the editor "dirty" — the **Save bar** appears at the bottom of the transcript pane with **Undo**, **Discard**, and **Save changes** buttons. Edits aren't persisted until you click **Save changes**.
### Karaoke highlight
While the audio plays, the word currently being spoken is highlighted in amber. The highlight follows along automatically — no setup needed. The estimate is based on the message's start time and the length of the current message; over time, as our transcription engine gets more confident, the highlight will land on the exact spoken word.
## Top toolbar
A row of actions sits between the file header and the transcript:
| Action | Enables when | What it does |
| ---------------- | ------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Merge** | 2 or more *consecutive* lines are selected | Combines the selected lines into a single transcript bubble. Uses the first line's speaker and start time. The waveforms on the timeline re-bind to the merged bubble automatically. |
| **Split** | exactly 1 line is selected and it has at least 2 words | Splits the selected bubble in half (at the midpoint by word count). The left half keeps any waveforms attached; the right half starts unattached — drag waves onto it from the timeline to assign them. |
| **Reassign** | at least 1 line is selected | Opens a dropdown of speakers; pick a speaker to reassign every selected line to. The waveforms on the timeline change color in lockstep. |
| **Undo** | at least one change has been made | Reverts the last change. History is bounded at the last 50 changes. **Cmd + Z** (macOS) / **Ctrl + Z** (Windows / Linux) does the same thing — except when you're editing a word, where it falls through to the browser's native text-input undo. |
| **Save changes** | the editor is dirty | Persists your edits and clears the dirty flag. |
## Audio Timeline
A sticky strip at the bottom of the transcript pane. The timeline always renders left-to-right, even on Arabic pages, because timestamps and waveforms only make sense in that direction.
| Sub-component | What it does |
| ---------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Timeline Header** | A ruler with ticks. The interval auto-scales: a tick every **5 s** for files under 1 minute, every **15 s** for files under 2 minutes, every **30 s** for longer files. |
| **Timeline Row** (one per speaker) | An editable speaker name input on the start edge plus the speaker's waveform segments laid out by time. Each segment's width is proportional to its duration. |
| **Waveform** | Animated bars drawn for each segment. **Speaker 1** is blue. **Speaker 2** is green. Any segment that the engine flagged as "negative" (frustrated / problematic) renders red, regardless of which speaker it belongs to. The colors swap automatically when you toggle dark mode. |
| **Timeline Controls** | A primary **Play / Pause** circular button (40 px) and a small **speed** toggle that cycles **1.0x → 1.5x → 2.0x → 0.5x → 1.0x**. |
| **Timeline Scrubber** | Click anywhere on the timeline to jump the audio to that moment. The position indicator slides to where you clicked. |
### Drag a wave to fix a wrong speaker or wrong time
A single drag does both:
* **Drag down or up** to move the wave to a different speaker.
* **Drag left or right** to move the wave to a different time.
* **Both at once** is also supported — diagonal drags work.
While you drag, the segment dims to 30% opacity and the row underneath the cursor highlights with a dashed outline. Release the mouse / finger to commit. The corresponding transcript bubble updates immediately — its color and timestamp follow the wave.
A tap or click on a wave (no drag) seeks the playhead to that wave's start. The audio meter slides to that point.
The whole drag gesture counts as a single undo step — pressing **Undo** once returns the wave to where it was.
While the audio is still loading, the timeline shows a centered spinner with the label **Loading timeline...** so you don't see a misaligned ruler.
## Save bar
Whenever the editor has unsaved edits, a sliding bar appears below the audio timeline:
* A pulsing amber dot + **Unsaved changes** label so you can't miss it.
* **Undo** — reverts the last change.
* **Discard** — opens a confirmation dialog ("All edits since the last save will be lost. This can't be undone."). Confirming reverts everything since the last save.
* **Save changes** — persists everything and the bar disappears.
## Unsaved-changes guard
If you try to navigate away (click another sidebar item, the dashboard logo, your profile menu, etc.) while you have unsaved edits, a dialog appears with three options:
* **Save and continue** — saves your edits, then takes you to where you wanted to go.
* **Discard** (red) — drops your edits, then takes you to where you wanted to go.
* **Cancel** — keeps you on the page; nothing happens.
If you try to refresh the page or close the tab, your browser will show its own "Leave site?" prompt instead — that one is the browser's native dialog, so we can't customise the buttons.
## Right panel
### Transcribe Settings (top of the panel)
| Field | What it does |
| ---------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Language Detection** | An outline button showing the language detection setting you chose pre-processing (defaults to **Auto-Detect**). The Globe chip in the file header shows the actual detected language separately. |
| **Include subtitles** switch | Turn on to generate subtitles for export to SRT or VTT. The AI labels each speaker so you can rename them on the fly. |
| **Detect words** button | An outline button that runs word-level detection over the audio. |
| **Run spell check** button | A brand-gradient primary button that runs spell check across the transcript. |
### Tabs section (middle of the panel)
Two tabs: **Summary** and **Intent**.
* A row per speaker showing **Speaker 1** / **Speaker 2** label and the editable name.
* **Executive Overview** heading with a thin bordered underline, followed by the conversation's narrative paragraph.
* A 350 px tall canvas with two animated bubble groups: **positive** and **negative**.
* Each bubble's size corresponds to how often that intent shows up in the conversation.
* Sample intents: *good (5)*, *fine (2)*, *perfect (1)* on the positive side; *issue (2)*, *problem (1)* on the negative side.
* The bubbles bob and float gently to keep the visualisation alive.
### Cue Properties (bottom of the panel)
A static card showing the time bounds of the current audio:
| Field | Format | Example |
| ----------------- | ---------------------------------------------------------------------------- | ------------------------------- |
| **Start time** | `HH:MM:SS.cc` | `00:00:00.00` |
| **End time** | `HH:MM:SS.cc` | matches the live audio duration |
| **Delete** button | An outline button styled red (red text, red border). Removes the active cue. | |
## Common questions
A click without movement seeks. To drag, press the mouse / finger down on a wave and move it more than a few pixels. The cursor changes to "grab" while you're dragging.
The bubble follows the wave automatically. If it didn't, your audio timeline and your transcript may be looking at different files — refresh the page and try again. If it persists, file a bug.
Speaker names are saved when you click outside the input. Then click **Save changes** in the save bar at the bottom of the transcript pane (or in the top toolbar) to commit the change.
That's intentional. While you're typing in a word's input, Cmd / Ctrl + Z is the browser's native text-input undo. Click anywhere outside the input first, then Cmd / Ctrl + Z runs the editor's undo.
The unsaved-changes prompt should always appear when you click a sidebar item or the logo with unsaved edits. If it didn't appear, you may have used a button-driven shortcut that bypasses the guard — file a bug. To recover, the previous save state is gone.
Three specific errors are surfaced now: "Couldn't reach the audio file" means a network or CORS issue, "Audio file unavailable (HTTP X)" means the server returned an error code (X), and a generic message means something else went wrong. If it persists, refresh the page; if it still fails, the file may have expired — re-upload it.
PDF export opens the formatted transcript in a new browser tab. Use your browser's **File → Print → Save as PDF** to save it. We'll add direct `.pdf` download once the server-side export ships.
The current `.doc` is a Word-readable HTML document. Microsoft Word opens it natively. We'll switch to a real `.docx` once the server-side export ships.
Exports are capped at 5,000 messages and 1,000 words per message — a "… \[truncated for export]" notice marks where the cap kicked in, and a yellow toast appears alongside the success toast. Server-side export (which removes the cap) is on the roadmap.
Speaker 1 is blue, Speaker 2 is green. If you see one color, the file probably wasn't transcribed with **Detect Speakers Automatically** turned on. Re-upload through [Upload Audio](/en/Systems/Speech-to-Text/Upload-Audio) with that switch on.
Wait until the **Loading timeline...** spinner clears, controls don't engage until the audio's metadata has loaded. If it never clears, the audio file may not be available yet — refresh the page after a few seconds.
Click the **Language Detection** button at the top of the right panel and pick the right language. Re-running detection with the right language often improves the transcript quality on the next refresh.