Yes, you can make AI voices talk to each other for free: write a dialogue, give each speaker their own voice and SpeakScribe Multi-voice reads it as one conversation in a single MP3. A free account makes 3 dialogues a month with 2 speakers (up to 1,500 characters each) — no card needed — and you can write a dialogue and preview every voice without an account. Paid plans go up to 10 speakers and full-length scripts, from 361 voices in 75 languages.
Each file below was rendered by the same Multi-voice pipeline you use in Voice Studio on
November 4, 2026, with no editing afterwards. The script is shown under every player so you can follow who says what.
Podcast intro
English
Two Global voices from the United States trading short questions and answers.
Pick voices that are easy to tell apart — a female and a male voice, or two different accents. Press ▶ to hear each one; previews don't use a render.
Write the conversation line by line and choose who says each line, or paste a script with names (Asha: … / Ravi: …).
Press Render dialogue. You get one MP3 with both voices, the same loudness and a short pause between speakers.
Two-voice text to speech is included in the free account (3 dialogues a month).
A free AI podcast with two hosts
A two-host podcast intro or short episode is a two-speaker dialogue: one voice
asks, the other explains. Keep each line to one or two sentences, open with a short greeting from the host, and let the
guest give the longer answers — that rhythm sounds more natural than two long monologues.
On a free account a dialogue can be up to 1,500 characters,
enough for a 1–2 minute intro or trailer. A full episode (10–20 minutes is roughly 10,000–20,000 characters) needs a paid
plan: Starter renders up to 30,000 characters in one dialogue.
Open Voice Studio, switch to Multi-voice and give each speaker a name. Every speaker gets a voice you can change and preview for free.
2
Write who says what
Type each line and pick who says it, or paste a whole script with the speaker's name before each line (Asha: …, Ravi: …) and SpeakScribe splits it into lines for you.
3
Render one file
Press Render dialogue. Each line is read by its speaker's voice, the volume is evened out between voices, short pauses are added between speakers and you download a single MP3.
No special format to learn: you choose the speaker from a list next to
each line. If you already have a script, paste it with names in front of the lines and it becomes one line per speaker.
A line without a name continues the previous speaker's line; text before the first name is given to a Narrator.
What happens to your dialogue — measured
Every line is rendered separately by its speaker's voice, brought to
the same average loudness and joined with a 0.35-second pause when the speaker changes (0.15 seconds when the same
speaker continues). Global voices come from Microsoft's neural voice service; Studio voices run on our own servers.
Here is what that took for the sample files (November 4, 2026, a 2-CPU worker):
Dialogue
Lines
Characters
Audio
Render time
Loudness gap before → after
Podcast intro
6
478
32.5 s
35.1 s
6.2 dB → 0.3 dB
Short story in Hindi
5
234
25.4 s
26.2 s
7.7 dB → 0.0 dB
Spanish lesson in two languages
6
326
30.1 s
28.8 s
4.4 dB → 0.4 dB
Kids' science Q&A
6
398
27.6 s
22.7 s
1.5 dB → 0.0 dB
Load test, 6 speakers
30
3,120
197.2 s
142.4 s
6.7 dB → 0.2 dB
Loudness gap = the difference in average loudness between the loudest and the
quietest line of the file. A gap of 6–8 dB is the "one person shouts, the other whispers" effect you get when you
join clips from different voices by hand. Render time grows with the number of lines (each line is its own request);
a short two-person dialogue is usually ready in under a minute.
What people make with several voices
Stories and audio drama
A narrator on a calm voice and one voice per character. Keep the
narrator on a different voice family (Studio vs Global) or gender from the characters so listeners can follow without visuals.
Podcast-style intros and interviews
Host and guest trading questions. Short questions and longer
answers sound most natural; split answers longer than three sentences into two lines.
Language lessons
A teacher in the learner's language and a native speaker in the
language being taught — each line read by a voice that actually speaks that language, in one file.
Kids' Q&A and quizzes
A child asks, an adult answers. Pair a lighter voice with a
deeper one for contrast — the sample above uses Bella (Studio) with Guy (Global).
Explainers with two presenters
One voice states the idea, the other asks the question the
viewer would ask — a simple way to break up a long explainer without showing a presenter.
Customer conversations and training
Role-play a call between a customer and an agent for onboarding
material. Subtitles with speaker names make the file easy to review.
For long-form work
Audiobooks and stories with multiple narrators
Paid plans are built for full chapters and client work: a narrator plus a
voice for every character, rendered in one pass and joined into a single file.
Speakers
Up to 10 voices in one dialogue
Length
Up to 150,000 characters per dialogue
Subtitles
SRT & VTT with the speaker's name
Long jobs
Rendered line by line — leave the page, find it in your history
Audiobooks and audio drama: keep the narrator on one voice for the whole book and give each recurring character the same voice in every chapter.
E-learning and training: role-play a customer call, a doctor–patient conversation or a job interview, with subtitles learners can follow.
Language courses: a teacher in the learner's language and native speakers in the language being taught, in one file.
Pick voices that contrast. Different gender, accent or pitch makes it obvious who is
speaking. Two similar voices in a row are hard to tell apart without video.
Write lines the way people talk. Contractions, short sentences and a question mark at the
end of questions change how the voice reads the line.
Keep each line in its voice's language. A Hindi voice reads Hindi; give an English line to
an English voice. Mixing languages inside one line makes the voice guess.
Use the speaker names for subtitles. The names you type appear in the subtitle files, so
use the names your viewers should see.
Preview before you render. Every speaker card has a play button that plays a ready-made
sample of that voice without using a render.
Two-person YouTube videos and podcasts
A structure that works for a 5–8 minute video
Cold open (1–2 lines): the guest says the most surprising fact first.
Host intro (1 line): who they are and what the episode answers.
Three segments: each starts with a host question and ends with a one-line takeaway; a new video chapter per segment.
Wrap-up (2–3 lines): the host summarises, the guest gives one action viewers can take.
Maya: Most people record their podcast too far from the microphone.
Sam: Welcome to Studio Notes. Today: three cheap fixes for better audio.
Sam: First fix. How far away should you be?
Maya: About a hand's width. Any further and the room starts to echo.
Use the names your viewers should see: the speaker name you type is the one that appears in the subtitles.
What you get back
One MP3 with every line in its speaker's voice and each line levelled to the same average loudness, so you can drop it straight onto the timeline.
SRT and VTT subtitles on paid plans. The first subtitle of each turn starts with the speaker's name, for example Maya: Most people record…, which YouTube shows as-is.
One render from your plan per dialogue, however many speakers it has.
For a video, put a still image or simple waveform of each host on screen while they speak and add chapters at each segment. For a podcast feed, export the MP3 as it is. Choosing two voices: pick a pair far apart in pitch. Some voice pages already suggest the most contrasting partner, for example Sonia or Brian.
Multi-voice limits by plan
A dialogue uses one render, whatever the number of lines. A free account can make
3 dialogues a month with 2 speakers and up to 1,500 characters each;
writing a dialogue and previewing voices needs no account. Single-voice text to speech stays free with 10 renders a month.
Yes, to try: a free account can render 3 dialogues a month with 2 speakers and up to 1,500 characters each (they count towards the free monthly renders). Writing a dialogue and previewing voices needs no account. Paid plans remove the monthly dialogue limit and allow more speakers (Starter 4, Pro 6, Studio 8, Agency 10) and full-length dialogues.
Open Voice Studio, switch to Multi-voice and give each of the two speakers a voice. Write the conversation line by line, choosing who says each line (or paste a script with names such as "Asha: …"), then press Render dialogue. You get one MP3 with both voices at the same loudness and a short pause between speakers. A free account can do this 3 times a month.
Yes, for short pieces: a free dialogue can be up to 1,500 characters, about 1–2 minutes — enough for an intro, trailer or short segment with a host and a guest. Full episodes need a paid plan (Starter renders up to 30,000 characters in one dialogue).
Yes. Every speaker in a dialogue has their own voice — 2 on the free account and up to 10 on Agency, so a narrator and several characters fit in one file. Keep each character on the same voice across chapters so listeners recognise them.
Up to 4 speakers on Starter, 6 on Pro, 8 on Studio and 10 on Agency. Any speaker can use any of the 361 voices, and two speakers can share a voice.
Yes. Each line is read by its speaker's own voice, so a dialogue can mix languages — the Spanish lesson sample on this page has a British English teacher and a Mexican Spanish speaker in one file. Write each line in the language of the voice that reads it.
Yes. Different voices are recorded at different loudness — in our test files the gap between the loudest and the quietest speaker was up to 7.7 dB before levelling. SpeakScribe levels every line to the same average loudness, which brought the gap down to 0.4 dB or less (measured November 4, 2026).
A dialogue counts as one render and can be as long as your plan's per-render limit: Starter 30,000, Pro 70,000, Studio 100,000, Agency 150,000 characters. Long dialogues are rendered line by line on our servers and you can leave the page while it works; the finished MP3 appears in your history.
Yes, on paid plans. Every dialogue also comes with SRT and VTT subtitle files timed to the voices, and the first subtitle of each line starts with the speaker's name, for example "Asha: Are you still waiting?".
More speakers (Starter 4, Pro 6, Studio 8, Agency 10), full-length dialogues (up to 150,000 characters on Agency), no monthly dialogue limit, subtitles with speaker names and longer file storage. The free account keeps 3 dialogues a month with 2 speakers.