Dialogue — Two Voices, One Recording

October 9, 2026

Ask for a conversation and Alfrada OS voices every line, then hands you one audio file with a clock time for each line. Alfred and Ada use their own voices. Anyone else uses an ElevenLabs voice you already have. Put each person on screen for those times and that file is the voice.

What you can do

• Voice a whole script in one call. Pass the lines in order. Each line is one turn: who speaks, then the words.
• Use Alfred or Ada by name. A line that names them uses that presenter's registered voice. A colleague you have registered is the same, once their voice id is in the presenter folder.
• Bring in any other ElevenLabs voice. Name the speaker and pass that voice's id. The id has to already exist in ElevenLabs. This does not invent a voice from a photo or a recording.
• Set the language for the whole conversation. Hindi is hi. English is en. Every line in that call is spoken in that language.
• Shape a line with a tag. Put [laughs], [whispers], [excited], [sighs], or [sarcastic] in the words where you want that delivery.
• The recording is Eleven v4. Dialogue uses the same expressive model as Speech. You do not name the model.
• Long scripts are joined for you. A request stays within 10 voices and about 2,000 characters. A longer script is split on sentence ends, generated in order on Eleven v4, and joined into one MP3.
• Each line comes back with a start and an end. The timing file sits next to the MP3, so a later edit can place a line on a video timeline. Lay the MP3 under a picture with the Video Editor.
• Put a face on each line. One presenter clip per speaker, 4–15 seconds, with that line in the prompt. Their registered portrait supplies the face. Alfred, Ada, or anyone in your presenter folder can be on camera.
• Hear this recording on that video. Strip the clips' own audio and mix the Dialogue MP3 under the picture. Say that in the same prompt. The agent runs the mix. Each person is on screen for their start and end. A short clip holds its last frame. A gap between lines shows no face. Overlapping times are turned away.

Where this shows up

You want Alfred to open and Ada to answer, without recording two files and lining them up by hand. Ask for the script. One MP3 comes back, Ada's line starting where Alfred's ends.

You want a guest in the conversation, an older voice that is not Alfred. Pass that ElevenLabs voice id on the guest's lines and leave Alfred and Ada as names.

You want the same exchange in Hindi. Set the language to hi and write the lines in Hindi. The voices stay the ones you named.

You want Alfred to open and Ada to answer, on camera, in the voices Dialogue just made. The timing file places each face, and the MP3 is the voice. The recipe is in Creative Studio.

Try it

• "Use Dialogue. Alfred: Welcome back. Today we are looking at the dashboard. Ada: And I'm Ada. Here's the number that changed. Alfred: It is up twelve percent since Monday."
• "Use Dialogue in Hindi. Grandfather, voice id 21m00Tcm4TlvDq8ikWAM: नमस्ते। आज मैं आपको यह कहानी सुनाता हूँ। Ada: And I'm Ada. Tell me what happens next."
• "Voice this two-host script with Dialogue, then mix the MP3 under in/demo.mp4 as narration."
• "Use Dialogue. Alfred: [calm] Welcome back. Today we are looking at the dashboard. Ada: [excited] And I'm Ada. The total is up twelve percent. Make a presenter clip of each of them saying their line. Put each one on screen for their timestamps, over in/demo.mp4. Strip the clips' own audio and mix the Dialogue MP3 under the picture as the voice."

Heads up

• Pro and above. Dialogue is in the same band as Speech. Community does not include it.
• One call is 500,000 tokens, the same as one Speech call, whether the script is short or long enough to be split.
• A voice id is not a description, and it is not a face. "An old man" does not select a voice. Alfred and Ada can be named. Everyone else needs the id from your ElevenLabs voice library for the soundtrack, and a registered presenter if they should appear on camera.
• Presenter Mode needs a screen recording, or one avatar track, before the speaker lane runs. Each presenter clip is one Video Lab generation. The export is a Video Editor render. Both are Pro and above, the same as Dialogue.