Speaker diarization tool
label every speaker automatically
Automatically identify and label every speaker in your audio and video. EditAssist separates overlapping voices, assigns speaker labels, and marks the timeline with who spoke when. Runs locally on your own machine.
Workflow
Ingest your audio or video and the agent separates every speaker using local models. It handles overlapping dialogue, cross-talk, and varying audio quality. Each speaker gets a consistent label across the entire project, and the timeline is marked with speaker segments ready for editing.
Use the labels to build speaker-specific selects, remove off-mic contributions, or normalise levels per person. Export speaker transcripts as separate files, or burn speaker-attributed captions into the video. The diarization runs entirely on your GPU with no audio uploaded.
Prompts
“Identify every speaker in the podcast and label the timeline markers with their names.”
“Label which guest is speaking in each segment.”
“Auto-label all speakers in this four-guest roundtable.”
“Find all clips where dialogue overlaps between two speakers.”
“Match the audio levels between the lapel mic and the boom across the entire scene.”
“Separate dialogue, music, and effects into stems for the mix.”
“Find every clip where there's laughter in the background.”
“Detect every time I talk over my guest and flag it.”
Features
Local speaker separation
Separates and labels speakers using local models. No audio is uploaded. Handles overlapping dialogue and cross-talk.
Consistent labels
Assigns the same speaker label across the entire project, even when the same person appears in different files or sessions.
Timeline markers
Adds speaker segments to the timeline as markers or colour-coded regions, ready for editing and export.
Per-speaker export
Export speaker-specific transcripts, stems, or captions. Normalise levels per speaker for consistent mix.
Local-first
All transcription, vision analysis, semantic search, and clustering run on your own GPU. Nothing about your footage is uploaded. The only data that leaves is the text of your conversation with the agent. It's free to download, local models are free, and a new account starts with £15 of credit and no card. For long-running cloud work, pay as you go, or choose £20/mo Starter or £100/mo Pro.
Getting started
- 01Download EditAssist for macOS or Windows and create an account. You get £15 credit and no card is needed.
- 02Ingest your footage. The agent transcribes, indexes, and clusters automatically.
- 03Describe what you want in plain English and let the agent build the cut.
- 04Review, refine with conversational commands, and export to your delivery spec.
More tools: AI transcript editor, AI auto captions, Podcast to Shorts. Or see everything EditAssist connects to.