Skip to main content

Search & Filters

OpenTranscribe provides powerful hybrid search combining full-text and semantic search with advanced filtering.

Search transcripts page with advanced filters sidebar

The Search Page​

Click Transcript Search in the top navigation bar to open the Search Transcripts page. Type a query and press Enter (or click the magnifier button); suggestions appear as you type, and choosing a title suggestion opens that file directly.

Result sources​

Once a query has run, a row of chips lets you choose where to search: Title, Transcript, Summary and Speaker. All four are on by default, each chip shows its match count, and you can turn any of them off. The chip counts show keyword matches only; in Smart mode additional related results can appear. If you turn every chip off, the page asks you to select at least one source. Transcript, title and speaker hits are listed under Transcripts, and summary hits under Summaries (each section can be collapsed).

Smart and Exact modes​

Above the results, a toggle switches between:

  • Smart (default): semantic plus keyword search. Related passages that do not contain your exact words are marked Related.
  • Exact: keyword-only search for exact matches.

A status indicator beside the toggle reads Semantic on or Keyword only. If it reads Keyword only in Smart mode, the neural model is not deployed on that server and the search is falling back to keywords (see Neural Search Setup).

Sorting​

Sort by offers Relevance (default), Upload Date, Completed Date, Filename, Duration and File Size. Everything except Relevance can be ascending or descending.

Result cards and Details​

Each result is one file: its title, a line of date, duration and relevance, and snippets with the matching words highlighted. A file with several matches lists the first few and offers Show N more matches; Collapse all / Expand all above the list opens or closes every match list at once.

  • The play button on a snippet previews the audio or video from that point in a floating player.
  • View Transcript opens the transcript with the matches highlighted, and View summary does the same for a summary hit.
  • Details (Pin details on wide screens) shows the file's speakers, tags, collections, size and the sources that matched. On a wide screen the details sit in a side panel that follows the card you hover or focus; pinning keeps it on one result. Clicking a tag or collection in the details narrows the results to it. Open file goes to the file page.

Speaker operator​

Type speaker:"Name" (for example speaker:"Joe Rogan" china) to search only within what that speaker said.

Semantic search finds conceptually similar content even when the exact words differ, and works across paraphrases. In Smart mode it runs alongside keyword search and the two result lists are fused into one ranking. If the neural model is unavailable, search falls back to keywords automatically.

Hybrid Search Fix (v0.4.0)

A critical bug in OpenSearch 3.4 caused hybrid search to silently fall back to keyword-only (BM25) search due to an ArrayIndexOutOfBoundsException triggered by the combination of aggregations, collapse, and RRF search pipelines. This was fixed in v0.4.0 — semantic search now works as intended. The improvement is significant: for example, searching "pytorch" went from 1 result to 83 results after the fix. Typo tolerance and query latency (255–500ms) are also confirmed working. If you were using an earlier version, reindexing is not required — the fix is in the query layer.

Filters​

The filter sidebar (on the Search page and in the gallery, where it can be collapsed and reopened with Show filters panel) narrows results by:

  • File Type: audio vs video
  • Speakers: files with a specific speaker (named speakers are listed individually; speakers nobody has named yet are grouped under Unlabeled speakers)
  • Duration: file length
  • Date Range: upload date
  • Processing Status
  • Tags: custom labels
  • Collections: grouped media files
  • Language: shown when your library has more than one language
  • Ownership (gallery only): All Files, My Files or Shared with Me, plus an owner picker

Reset clears every filter. In the gallery, the filename and title search box is in the toolbar above the file list rather than in the sidebar.

Gallery overview with filter sidebar showing tags, collections, and speakers

The main file gallery supports multiple display modes:

  • Grid View: Card-based layout showing thumbnails and metadata (default)
  • List View: Compact table layout for faster scanning of large libraries
  • Sorting: the Sort by menu offers Upload Date, Completed Date, Filename, Duration and File Size
  • Infinite scroll: more files load as you scroll, and a count shows how many of the matching files are loaded
  • Virtual Scrolling: Smooth performance even with thousands of files

Search Query Syntax​

Keyword Search Behavior​

Keyword search is a plain-language search; it does not parse Boolean operators. Words such as AND, OR and NOT, and wildcard characters such as *, are not special and are treated as ordinary text.

SyntaxExampleDescription
Several wordsproject deadlineMatches transcripts containing the words, ranking passages where they appear close together or in order higher
Exact phrase"project deadline"Matches the exact phrase, with no typo tolerance
Typo toleranceAutomaticHandles typos and minor spelling variations (unquoted queries only)
Speakerspeaker:"Joe Rogan" chinaSearches only that speaker's words

Semantic Search Tips​

Semantic search finds conceptually related content even when exact words differ:

  • Search for "budget concerns" to find mentions of "financial constraints", "cost overruns", or "spending limits"
  • Search for "next steps" to find action items, follow-ups, and task assignments
  • Works best with natural language queries rather than single keywords
  • Semantic search runs automatically alongside keyword search in Smart mode

For maximum precision, combine text queries with filters:

  1. Enter your search query in the search bar
  2. Use the filter sidebar (Show filters panel if it is collapsed)
  3. Apply filters (speakers, date range, tags, collections, file type, and so on)
  4. Results are narrowed to match both the query and all active filters; a Filters applied to results note confirms they are active

Search Result Highlighting​

When search results are returned, the matching text is highlighted in the result snippets, so you can see exactly where your search terms appear within each transcript. Words found only by semantic similarity are highlighted differently from exact keyword matches.

Search results with highlighted matching transcript segments

Finding Text Inside a Transcript​

The gallery/library search above searches across your whole collection. A separate find bar searches within the transcript or summary you currently have open, and shares its component and fuzzy-matching behavior (diacritic- and case-folding) everywhere it appears, so it behaves the same way in the transcript viewer as it does in the summary panel.

In the transcript viewer, the find bar highlights matches in the currently loaded portion of the transcript instantly, then resolves a lightweight, debounced count query scoped to that file to show an accurate N of M+ indicator and progressively load more matches if any exist outside what's currently rendered — so a match late in a long transcript is never missed just because that page hasn't loaded yet. In the summary panel, where the whole summary is already in memory, the find bar is a simpler, complete-from-the-start version of the same component.

Searching Settings​

Settings has its own macOS-style search: a search box above the sidebar tabs that replaces the grouped navigation with ranked, highlighted results as you type. Selecting a result jumps straight to that section and briefly flashes the matched control, so you don't have to remember which tab a setting lives under. It works across all supported languages and only surfaces settings your account and edition can actually see.

Transcript Comments​

Add timestamped comments to any transcript for notes, annotations, or collaboration.

Adding a Comment​

  1. Open a transcript and navigate to the Comments section
  2. Type your comment in the text area
  3. Click "Mark Current Time" to attach the current playback timestamp -- a timestamp is required
  4. Click "Add Comment" to save

Timestamp-Linked Comments​

Every comment is linked to a specific point in the media file:

  • Comments display the timestamp in MM:SS format
  • Click the timestamp badge on any comment to jump the player to that moment
  • Comments are sorted chronologically by timestamp
  • You can clear and re-set the timestamp before submitting

Editing and Deleting Comments​

  • Edit: Click the Edit link on your own comment, modify the text, then click Save
  • Delete: Click the Delete link on your own comment and confirm the deletion
  • Only the comment author can edit or delete their comments
  • A confirmation dialog appears before deletion to prevent accidental removal

Comments in Exports​

When exporting a transcript as .txt, you can optionally include comments in the export output (see Export Options).

Next Steps​