1
Add your audio
A folder can be renamed in place
Drop audio files here
mp3, wav, m4a, ogg, flac, opus
- Files
- 0
- Total length
- —
- Source
- not chosen
- Transcribed
- 0
Batch rename voice-over files based on what is actually said in them
Recording a game's dialogue leaves you with a folder of takes called take_01.wav, ZOOM0043.mp3 or Untitled Session 7.m4a. Renaming each one means opening it, listening, finding the matching line in the script, and typing the ID by hand. For a mission with a hundred and fifty lines that is most of a day.
This tool transcribes every recording with Whisper, compares each transcript against the dialogue in your script, and renames the file to the line it belongs to. On a real 147-line table with transcripts degraded to a 20% word error rate, it placed 96% of takes correctly. Everything runs in the browser, so no audio is uploaded and there is nothing to install.
How it works
- Add your audio. Drop files in, or pick a whole folder so files can be renamed in place rather than downloaded as copies.
- Load your dialogue script. Any CSV works. Column roles are detected automatically and can be remapped by hand. If you do not have one yet, download a starter template in one of four layouts.
- Transcribe. Whisper runs locally on your CPU or graphics card. The model downloads once and is cached afterwards.
- Match and rename. Each take is paired with a line, best match first, so two takes can never claim the same line. Weak matches are flagged and a dropdown lets you correct them.
- Write it back. Rename in the folder, download renamed copies, or export an updated CSV with the audio asset paths filled in.
What else it does
Audio format conversion
Convert between formats on the same pass, including MP3 to WAV, M4A to WAV, and WAV to MP3, with control over sample rate, channel count, bit depth and bitrate. Decoding uses the browser's own audio engine, so mp3, wav, m4a, aac, ogg, opus, flac, webm and the audio track of mp4 files are all accepted.
Voice effects for games
Sixteen presets built from Web Audio filters: telephone, radio chatter, walkie-talkie, megaphone, PA system, intercom, robot, alien, monster, ghost, echo, reverb, cave, underwater, static and vinyl. Each one is level-matched to the source so nothing clips or comes out faint, and a preview button plays six seconds before you commit.
Unreal Engine data tables
Load an EventAudioManager data table and the tool fills the AudioAsset column with a full object path and updates AudioKey to match the new filename. Names are restricted to characters Unreal accepts in asset names, and the case authored in your table is preserved. Every other column is left untouched.
Questions
Is my audio uploaded anywhere?
No. In local mode the transcription model runs inside your browser and the audio never leaves your machine. The only network requests are for the model weights themselves, which are fetched once from a public CDN and then cached. There is an optional API mode that sends audio to an endpoint you configure with your own key, but it is off by default.
Which audio formats are supported?
Anything your browser can decode, which covers mp3, wav, m4a, aac, ogg, opus, flac, webm and the audio track of mp4 and mov files. Output can be WAV at 16-bit, 24-bit or 32-bit float, or MP3 between 64 and 320 kbps.
Can it rename files directly in a folder?
Yes, in Chrome and Edge, which implement the File System Access API. Firefox and Safari cannot rename files on disk, so there you transcribe and download renamed copies instead, or export a rename script that renames the originals when you run it.
How accurate is the script matching?
Measured against a real 147-line dialogue table with transcripts artificially degraded, it placed 96% of takes correctly at a 20% word error rate and 95% at 30%. Short lines like "Copy." and "Say again." matched perfectly, and audio belonging to no line in the table matched nothing rather than guessing. Every row shows its confidence score and a dropdown for correcting a weak match.
Does it work offline?
After the first run, largely yes. The Whisper model is cached by the browser, so transcription, conversion, effects and renaming all work without a connection. The page itself still has to be served from a web server rather than opened as a file.
Do I need a dialogue script to use it?
No. Without one you can still name files from the transcript itself, using tokens like the first five spoken words, keywords, the recording date, the duration, or blanks you fill in per file. The script is what lets it assign a specific line ID.
What format does the dialogue script need to be in?
Any CSV with a column of dialogue text. Column roles are detected from the names commonly used in game and film pipelines, including SubtitleText, Line, Dialogue, Text, Speaker, Character and Filename, and four dropdowns let you remap them if the guess is wrong. Four starter templates are available to download: a full EventAudioManager data table, a minimal version, a plain ID and Speaker and Line sheet, and a middleware layout.
How long does transcription take?
On a graphics card with WebGPU it is comfortably faster than real time. On CPU it is slower than real time, so a ten-minute recording can take fifteen to twenty minutes with the recommended model. Smaller models trade accuracy for speed, and the first run additionally downloads the model.
Is it free?
Yes, and there is no account, no upload and no usage limit, because the work happens on your own machine rather than on a server. It is made by Neutron Studios alongside their games.