WHO SAID WHAT
Speaker diarization, without the setup.
Run WhisperX on your conversation, review who said what, and name the speakers before you export.
WhisperX powered · Independent, unofficial service
Two-voice conversation
Play it. Change a word. Make it yours.
Result details & original output
THREE SIMPLE STEPS
How speaker diarization works
Upload your interview or conversation.
Run with speaker diarization enabled.
Review speaker turns, rename labels, and download.
A useful starting point for interviews
Speaker diarization groups speech into labels such as Speaker 1 and Speaker 2. It tells you when each voice appears within this recording; it does not identify a person or match voices across files. Name the speakers after listening to a few turns.
0.09s
How do I run WhisperX without installing it?
3.20s
Upload an audio file, choose your language, and start transcription.
8.27s
Can I download Word timestamps and subtitles?
Review the moments where voices overlap
Interruptions, background voices, similar sounding participants, and changes in microphone position can confuse speaker clustering. Listen at handoffs and correct labels before sharing the transcript. An automatic speaker count is a starting point, not a reliable headcount.
- Check the first clear turn for each participant.
- Listen to short interjections and overlapping speech.
- Keep an unknown label when the recording does not support an attribution.
Keep the speaker information you need
Use a speaker labeled text transcript for interview notes and JSON when your workflow needs individual timing data. A subtitle file is more useful when the conversation will accompany a video. Editing a speaker name changes its display label; it does not run the model again.
Read the speaker diarization guide ↗CHECK THE OUTPUT YOURSELF
Listen to the voices behind the result.
This demonstration uses a synthetic recording and actual WhisperX processing. Download the source and the original result to check what the engine returned.
Two-voice conversation · Original model result
Original script; two eSpeak NG voices; actual WhisperX and pyannote community-1 speaker labels.
Model: large-v3 · Provider: replicate
Pinned runtime version
655845d6190ef70573c669245f245892cd039df4b880a1e3a65852c09252f5ccTRANSCRIPT EXCERPT
0.09s · Voice 01How do I run WhisperX without installing it?
3.20s · Voice 00Upload an audio file, choose your language, and start transcription.
8.27s · Voice 00Can I download Word timestamps and subtitles?
Original output is separate from edits made in the interactive sample. Word times may include estimated positions; check them before precise timing work.
BEFORE YOU START
Prepare the file. Know what you will get.
Supported recordings and limits
Upload MP3, WAV, M4A, FLAC, OGG, MP4, MOV, or WebM. This hosted release supports English speech. Use a file you have permission to process; external links and live recording are not supported.
The one-time trial includes 5 processing minutes, with files up to 100 MB and 5 minutes long. Paid accounts support up to 500 MB and 60 minutes per file. Confirm the measured duration before starting; a failed task releases its reserved minutes. See the minute packs.
Choose the right next step
For a written draft from an audio recording, use audio to text. To review speech beside its original video, use video to text. For caption import, use the subtitle generator.
All workflows use the same WhisperX engine and account. Your results are private, with 7-day trial retention and 30-day paid retention. Download what you need to keep. Read the workflow guide and recording privacy information before your first upload.
GOOD TO KNOW
Speaker diarization questions
Does diarization know the speakers’ names?
No. The model assigns labels within a recording. You can rename those labels after checking the audio; the service does not identify people from a voice.
Does speaker diarization cost extra?
No. It is included in the minute price. The confirmed media duration determines the charge.
Will it separate two people speaking at once?
Overlapping speech can cause missing words or incorrect labels. This is a transcript workflow, not a tool that produces separate audio tracks for every participant.
LESS SETUP. MORE DONE.
Your next transcript starts here.
Try the public sample, then bring your own recording. Your first account includes 5 minutes to try the workflow.
Run WhisperX