Automatic Dialogue Editor
- Distinguish takes from pauses — a breath or hesitation within a take won’t cause a false split
- Detect retakes — recognizes when the speaker starts the same line over, even without a long silence
- Handle meta-commentary — filters out things like “let me try that again” between takes
- Classify take quality — marks each take as good (complete reading) or incomplete (cut off or interrupted), color-coding incomplete takes for quick visual identification
Analysis Modes
When you ask Reagent to analyze dialogue, it will ask you to choose between two modes:- Voice-takes — For studio recordings where a speaker reads lines with multiple takes per line. Reagent identifies individual takes, detects retakes, and classifies quality.
- Podcast — For long-form recordings with multiple speakers. Reagent aligns your audio against a reference script and identifies sections to keep or cut.
Podcast Mode
Podcast mode is designed for editing long-form recordings like podcasts, audio books, interviews, or narration sessions. When using podcast mode with a reference script, Reagent:- Aligns audio to your script — matches spoken sections to their corresponding lines in the script
- Shows suggested edits — gaps and off-script sections appear as red (incomplete) items in REAPER, so you can preview what will be cut before committing
Pipeline Stages
Under the hood, dialogue editing runs through three stages. You can let Reagent handle all three automatically, or run them individually for more control:- Transcribe — Listens to your voice recordings and generates word-level transcriptions with timestamps. Results are stored on each item for use in the next stages.
- Analyze — Reads the transcriptions and detects take boundaries, computes trim points, and classifies each take as good or incomplete. You can also provide custom analysis instructions instead of standard take detection — for example, finding all speaker names, marking questions vs. statements, or identifying topic changes.
- Apply Edits — Splits items at the detected take boundaries, trims silence, and renames each take to match the spoken text.
Script Reference Files
If you have a script, cue sheet, or line list for your recording, you can provide it as a reference file to improve take classification accuracy. Reagent compares the transcribed audio against your script to better identify which takes are complete readings and which are partial or off-script. Supported formats: CSV, TSV, TXT, XLSX, PDF, DOCX You can attach reference files directly in the chat input using the + button or by pasting the file path:Rename from Text
By default, edited media items will be renamed to match what was actually said. This makes it easy to identify takes at a glance in REAPER.Volume Leveling
Automatically level dialogue or vocal recordings by generating a take volume envelope that normalizes dynamic range. Works on any voice, dialogue, or vocal items.| Parameter | Default | What it controls |
|---|---|---|
| Target level | -18 dB | If you want to use LUFS specifically say “lufs” in your prompt otherwise this defaults to peak loudness. |
| Gain range | -12 to +12 dB | Min/max gain adjustment — limits how much the leveler can cut or boost |
| Attack | 20 ms | How quickly gain increases — lower values react faster to quiet sections |
| Release | 80 ms | How quickly gain decreases — lower values react faster to loud sections |
| Lookahead | 50 ms | Shifts the envelope earlier to anticipate level changes |
| Window size | 50 ms | Analysis window — larger windows produce smoother, less reactive leveling |
| Smoothing passes | 2 | Number of smoothing passes on the envelope to reduce artifacts |