► BG3 DIALOGUE FINDER 3.0 ◄
► OVERVIEW
A desktop tool for searching and managing Baldur's Gate 3 dialogue audio files. Search 291,000+ dialogue entries by character, dialogue text, audio type, or filename - then copy the .wem files you need to a destination folder.
Complete rewrite with a new UI and rebuilt database.
► DOWNLOADS
| Files | Version | Type | Description |
|---|---|---|---|
 BG3 Dialogue Finder Tool | v3.0 | Main | Download and extract the exe file |
► Old Versions
► INSTALLATION
- Download and extract the zip file
- Run BG3-Dialogue-Finder.exe
- Set your source folder(s) containing .wem audio files
- The three you absolutely have to set can be found in your BG3MM UnpackedData folder and are:
- UnpackedData\SharedSounds\Public\Shared\Assets\Sound
- UnpackedData\SharedSounds\Public\SharedDev\Assets\Sound
- UnpackedData\Voice\Mods\Gustav\Localization\English\Soundbanks
- The three you absolutely have to set can be found in your BG3MM UnpackedData folder and are:
- Set a destination folder for copied files
- Search and copy!
► NEW UI
Complete visual overhaul. The old v2 interface was a basic table with row selection and single-column sorting. The new table is more like a spreadsheet.
Results table
- 7 columns - Filename, Dialogue, Character, Type, Event, Source, Status (v2 had 5)
- Cell selection - Click and drag to select individual cells or ranges, not just whole rows
- Compact rows - Smaller font and tighter spacing so you can see more results at once
- Alternating row colors - Easier to read across long rows
- Click any column header to sort by that column
- Shift+click another header to add a secondary sort (e.g., sort by Character, then by Type within each character)
- Ctrl+click to clear all sorting
- Headers show ▲/▼ arrows with position numbers when multi-sorting
Right-click any cell for quick actions:
- Copy Selection (or Ctrl+C)
- Copy Filename
- Copy Dialogue
- Copy Row(s) - copies full rows with column headers
- Select Entire Row
- Add filters - Click "+ Add Filter" to add more search rows. Each row has a text field and a dropdown to pick what you're searching (Dialogue, Character, Type, Filename, Source).
- Search for multiple things - Two filters with the same type finds either match (Character: Astarion OR Character: Shadowheart)
- Narrow your search - Combine different filter types to require all conditions (Character: Astarion AND Type: Combat)
► SOURCE CONFIDENCE
Every entry now tracks where its data came from. Individual cells are colored based on their source - a single row can have multiple colors if data comes from different sources.
The colors
- Game (green) - Extracted from VoiceMeta. 100% accurate.
- Pattern (yellow) - Character parsed from sound event name. Sometimes reliable.
- Fallback (red) - No entries will show up like this yet, but they will be for characters guessed from event string. May be wrong. Basically just throwing **** at the wall to TRY and guess the character.
- Override / Transcribed (blue) - Manually added, or auto-transcribed by OpenAI Whisper. May have errors.
Click the ? button next to the source legend to see detailed explanations with example rows for each confidence level. When in doubt, find a similarly-colored row in the helper to interpret your data.
► DATABASE
Completely rebuilt. No longer uses @NoComply's database at all.
The old database used in v1 and v2 was a flat table with four columns: filename, dialogue, character, type. No way to know if data was accurate. Inconsistent character names. No event info.
What's new
- Source tracking - Every entry has a confidence level. Know instantly if character attribution is reliable.
- Event names - Sound event names now stored (e.g., UniqueNPC_LOW_Halsin_Combat_Pain).
- Normalized schema - Separate tables for characters and audio types with foreign keys.
291,297 total entries (up from 290,257 in v2 - +1,040 entries, +0.4%)
The entry count is nearly the same, but entries now have source tracking and metadata:
- 185,970 game - VoiceMeta dialogue with subtitles (100% accurate)
- 99,409 transcribed - dialogue auto-transcribed by OpenAI Whisper for files missing from VoiceMeta
- 3,818 pattern - character parsed from event names like UniqueNPC_GLO_Halsin_Combat_Pain
- 2,100 override - manual additions and corrections
- 87,346 entries now have event names (was 0)
- 662 audio types properly categorized (was 115 inconsistent labels)
- 2,091 characters normalized (was 1,910 with duplicates like "Astarion" vs "ASTARION")
- 237,209 entries have dialogue text, 54,088 are audio-only (efforts, combat, etc.)
► OTHER
- Settings persist via Windows registry
► CHANGELOG
► Show Spoiler
► WHAT I NEED FROM YOU *******
Report everything you can. Data irregularities, stuff that should be manually overridden, everything. The more you guys report, the more the database improves.
Feel free to report any bugs or findings in this thread.
► CREDITS
- @NoComply - Original database and inspiration
- @orinEsque - For helping me with datamining
- @Lancaster - For testing v3 early and help finding data issues
