Dialogue Text to Speech

How dialogue text to speech works

A dialogue is the simplest script there is: two names, alternating lines. Start every line with the speaker, a colon and then what they say. AudioDrama assigns each speaker a different voice, reads the lines in order and returns one MP3 in which the exchange sounds like a conversation, not like two clips pasted together.

Here is a short café conversation, the kind of exchange used for English listening practice. It is 342 bytes once speaker tags are counted, so it needs at least 23 seconds of credit.

Customer [friendly]: Hi, could I get a medium latte, please?
Barista: Of course. For here or to go?
Customer: To go, please. And do you have anything without nuts?
Barista [helpful]: The lemon loaf is nut-free. Would you like a slice?
Customer: Yes, please. How much is that altogether?
Barista: That's six dollars fifty.

Notice what is missing: no stage directions and no narrator. For a plain dialogue you do not need them. Lines without a name would be read aloud by the Narrator voice, so leave out anything you do not want spoken.

Write dialogue for the ear

Text that reads well on a page often sounds stiff when spoken. A few habits make generated conversations sound more natural:

• Keep turns short. Real people rarely speak in paragraphs. Two sentences per turn is a good default; split a long speech into pieces with a reaction in between.
• Use contractions. “That's six fifty” sounds like a person; “That is six dollars and fifty cents” sounds like a form.
• Let punctuation set the pace. Commas, full stops and question marks shape the rhythm. Ellipses and dashes can slow or break a line, though the effect varies by sentence.
• Read it aloud first. If you stumble on a line, the voice may too. Rewrite until it comes out in one breath.
• Add a small direction when tone matters. A line like Barista [helpful]: … costs a few bytes and can change the delivery. Results vary, so listen and adjust.

An explainer with a host and a guest

Interview-style explainers are another common two-speaker format. This one uses an emotion direction and an emphasized word (the <emphasis> tag stresses a word). It is 357 bytes, about 24 seconds at minimum.

Host [warm, curious]: So why does bread rise at all?
Guest [enthusiastic]: Yeast eats the sugars in the dough and releases carbon dioxide. The gas gets trapped, and the dough puffs up.
Host: And the smell?
Guest [laughing]: That's the best part. The smell comes from <emphasis>lots</emphasis> of compounds that form while it bakes.

One rule worth knowing: only one bracketed direction can sit between the name and the colon, as in Guest [laughing]:. Extra directions go after the colon, inside the line, for example Guest: [laughing] [short pause] That's the best part.

What people use two-voice dialogue for

• Language listening practice. Ordering food, asking directions, a job interview. The built-in voices are English; other languages may work but pronunciation is not guaranteed.
• Explainer and training videos. A question-and-answer format keeps an explanation moving. Use a Host who asks what a beginner would ask.
• Podcast and video drafts. Hear a conversation before you record it yourself, or use the output as a rough guide track.
• Game and app prototypes. Placeholder exchanges between a player character and an NPC.
• Sales and support role-play. Short scripted exchanges for practising objections or answers.

If your scene has more than two people or a narrator, read the page on text to speech with multiple voices. For a performance with feeling, see text to speech with emotions.

Limits and cost

• Up to 6,000 characters per script, with a 9,000-byte cap per job once speaker tags are included. A long conversation should be split into parts.
• One speaker per line; no overlapping speech, no sound effects, no music.
• 2 free minutes on sign-up; afterwards $9 for 60 minutes or $29 for 240 minutes, bought once.
• Credits are counted by audio length or script length (1 second per 15 bytes of script text, including directions), whichever is greater. A short two-person exchange of a few hundred bytes uses well under a minute of credit. See pricing.