Expressive TTS
Expressive Text to Speech
Most text-to-speech tools ask you to pick an emotion from a list. Drama 3 takes a different route: you describe the performance in the script itself, in your own words, and the direction sits exactly where the delivery should change.
Real Drama 3 generations
Each clip below is a generation of the exact script shown next to it. Open the full script, copy it, or send it to the studio and direct your own version.
Holding it together
[barely holding it together] I'm fine. Really. [shaky breath] Go on without me. I'll catch up. [pause] I just need a minute.
Cold open narration
[low, measured, like a documentary narrator] The city went quiet at 3 a.m. [short pause] No sirens. No traffic. <slow>Just the rain.</slow>
Villain reveal
[amused, taking his time] You came all this way. [soft laugh] For a file. [harder, dropping the warmth] Then let me show you what is in it.
Late-night radio host
[smooth, close to the microphone] You're listening to the midnight hour. [sigh] If you're awake with me, <emphasis>you're not alone</emphasis>.
Training video, plain then pointed
[clear and friendly] Welcome aboard. Today we cover the basics. [short pause] [firmer] And then we cover the mistakes that cost money.
Fairy tale opening
[hushed, telling a secret] Once, in a village that no longer exists, [pause] a girl traded her name for a lantern. <whisper>This is her story.</whisper>
Support call, patience thinning
[professional, polite] I understand. [short pause] [slower, holding it together] I have explained that twice. [exhale] Let us try once more.
Audiobook line, tender
[gentle, unhurried] She kept the letter for forty years. [pause] [almost a whisper] She never read it again, but she never threw it away either.
Expressing emotion without an emotion picker
These are the ideas that make a line sound performed instead of read aloud.
Describe the performance, don't select a preset
A direction such as [impatient but trying not to show it] carries more information than any dropdown label, because it says what the speaker is doing with the line.
Write emotion where it changes
Directions sit inline, so a sentence can open light and close heavy without splitting the script or generating twice.
Use sounds and pauses as punctuation
A breath, a sigh, a laugh, or a pause can be placed exactly where a person would take it, instead of hoping the model adds one.
Keep control local
Paired tags such as <whisper>…</whisper> change only the words they wrap, so one phrase can drop to a whisper while the rest of the line stays at full voice.
From flat text to a performance
Three passes, each one short.
Start with plain text
Paste the copy and read it aloud in your head. Mark the words that carry the emotion before touching any tags.
Write one direction per shift
Add a direction where the tone changes, in plain language. Describe the intent and the listener, not just the mood.
Generate and compare
Listen for the moment the direction lands. If the line still sounds read, sharpen the wording before adding another tag.
Where expressive delivery matters
Audiobooks and stories
Let the narration follow the scene: calm in one paragraph, tense in the next, without a second narrator.
Video voice-overs
Match the read to the cut, including the quiet lines that should not sound like an announcer.
Game and app characters
Give each line the attitude the moment calls for and regenerate the takes that miss.
Ad and social reads
Set the hook, the turn, and the close inside one short script.
E-learning
Keep instructions neutral and warnings pointed, in the same voice.
Accessibility narration
Add expression where it helps comprehension, and keep the rest plain.
Good to know
Expressive delivery has practical limits; these are the ones that come from the pipeline.
Up to 2000 characters per generation
The limit covers the whole script, directions included.
Emotion is written, not selected
There is no emotion dropdown; the performance comes from the words you write.
One MP3 per generation
Each take returns a single audio file to preview and download.
No voice cloning
The studio generates speech from your script and directions; cloning a voice is not offered.
Expressive text to speech questions
Do I need to pick an emotion from a list?
No. Drama 3 has no emotion picker; you describe the performance in plain language inside the script, for example [relieved, finally letting go].
Which is better: a mood word or a description?
A description. [trying to sound calm while panicking] tells the model what the speaker is doing, which shapes the take more reliably than a single mood word.
Can I change emotion mid-sentence?
Yes. Directions sit inline, so the tone can shift inside one line without splitting the generation.
How many directions should one line have?
Start with one, listen, and add more only where the delivery actually changes. Stacked directions compete with each other.
Is the result stable between takes?
Drama 3 is a preview model and Fish Audio notes that results vary, so generating a few takes of an important line is normal practice.
Write the emotion in
Paste your script, describe how it should be performed, and download the take.