How to Use Fish Audio: Emotional Text to Speech Guide
Fish Audio text to speech guide: how to write bracket directions for Drama 3, pick a voice from the library, engine and speed. Ready prompts and fixes.
Fish Audio is text to speech you can direct: before a line you write in square brackets how it should be played — whispering, tired, with a smile — and the Drama 3 engine delivers it that way. In the IIshki studio, Fish Audio comes with two engines (Drama 3 and S2.1 Pro), voices from the open Fish Audio library, up to 5,000 characters per run, 1 to 4 takes at once, and MP3 or WAV output. Below: how to write directions, which voice and engine to pick, ready-made prompts and the most common mistakes.
What Fish Audio does and how it differs from other TTS
Most voiceover models read text evenly and guess emotion from punctuation. Fish Audio works differently: you write a note for the "actor" right inside the text, and the model performs it. That matters when delivery counts more than plain clarity: ads, characters, stories, dialogue in a video, audio drama.
What you get in the IIshki studio:
- Two engines. Drama 3 is expressive and follows directions best. S2.1 Pro is faster and cheaper, understands tags too, but plays it calmer.
- Voices from the Fish Audio library. Six ready defaults (five Russian voices and an English narrator) plus search across the whole library with filters. A public voice can be found by its ID or a fish.audio/m/… link.
- Language. English, Russian, Ukrainian and 13 more in the list, plus Auto-detect from the text.
- Speed: slow, normal, fast. Format: MP3 or WAV.
- Editor buttons: emotions, pauses, sounds, and for Drama 3 also delivery directions.
What it doesn't have: cloning your own voice and music generation — for songs use ElevenLabs Music.
Which model to pick for the job
Fish Audio isn't the only voice model in the studio. If delivery doesn't matter but steady diction on long text does, another model may fit better.
| Job | Model | Why |
|---|---|---|
| Ad, character, a line with attitude | Fish Audio, Drama 3 | acts from bracket directions, switches delivery mid-line |
| Long narration, lesson, draft | Fish Audio, S2.1 Pro | faster and cheaper, sounds even |
| Voiceover with a large curated voice set | ElevenLabs Voice | five engines, emotion and delivery tags on v3 and v4, 39 voices |
| Even voiceover with emotions and pauses from the editor | MiniMax Audio | several Speech generations, HD and Turbo |
| Talking video from a photo and your audio | OmniHuman 2, Kling Avatar v2 | sync lips to your audio file |
To compare them side by side, see the roundup of AI voiceover tools, plus detailed guides on MiniMax and ElevenLabs.
How to use Fish Audio: 5 steps
1. Open the model and paste your text
Go to Fish Audio in the studio and paste your text — up to 5,000 characters. Split a long script into logical chunks of 1–3 paragraphs: it's easier to redo one line than the whole file.
2. Pick a voice
Start with one of the defaults or open the library and filter by language, gender and style. Listen to the sample first: timbre matters more than any bracket direction. For a series of videos, pick one voice per character and don't switch it mid-project.
3. Keep Drama 3 or switch to S2.1 Pro
For emotional text — Drama 3 (the default). For long, even narration and test runs — S2.1 Pro.
4. Add directions, pauses and sounds
A direction goes in square brackets before the line; a sound or pause goes exactly where it should be heard. The editor's emotion, pause and sound buttons insert the same markup.
5. Lock the language and speed, run 2–4 takes
Set the language explicitly instead of Auto, especially if the text mixes languages or brand names. Keep speed at Normal and don't change it between parts of one video — otherwise the voice sounds like a different person across the cut. Run 2–4 takes at once: they differ in intonation, and the best one is easy to pick by ear.
How to write directions for Drama 3: a formula
A direction is a brief for an actor, not a magic tag. The working formula: [emotion + how to deliver] line [sound or pause in the right spot] continuation.
A few rules that make a real difference:
- Put the direction at the start of the line it applies to. A new direction mid-line changes the delivery from that point.
- Place sounds precisely. [laughs], [sigh], [pause] happen where they stand. After a laugh tag, write the laugh out — "ha-ha" — so it sounds natural.
- No more than two or three directions per line. An overloaded line sounds theatrical and uneven.
- Specify intensity: [slightly annoyed] and [very tired] work better than plain [annoyed].
- Don't put an emotion on every sentence. Contrast between plain lines and one acted line sounds more alive than nonstop expression.
Ready prompts for Fish Audio
Ad, warm delivery:
[warm, smiling] Hi! Today I'll show you how to make a breakfast you won't get tired of in five minutes. [pause] Spoiler: you need just three ingredients.
Story with intrigue:
[whispering, conspiratorial] Listen, don't tell anyone. [pause] We launch tomorrow. [laughs] Ha-ha, well, now you know.
Character, exhaustion:
[tired, slow] It's 3 a.m. and the report still isn't done. [sigh] Fine. [a bit brighter] One more cup of coffee and we push through.
Delivery shift mid-line:
[calm, businesslike] So, the plan for the quarter is simple. [suddenly excited] Wait, did they already approve it?!
Even narration (S2.1 Pro, no directions):
Step one: open the settings. Step two: choose Notifications and turn off what you don't need. It takes less than a minute.
Two-character dialogue — voice each line separately, one voice per character, then put them together in editing:
[mocking] Did you really think I wouldn't notice? — [confused, quietly] I… wanted it to be a surprise.
Common mistakes and how to fix them
| Mistake | Cause | Fix |
|---|---|---|
| The direction is read out loud | it's in round brackets or has no brackets | use square brackets only: [whispering] |
| Barely any emotion | S2.1 Pro selected or direction too vague | switch to Drama 3, add intensity: [very tired] |
| Foreign accent | Auto language with mixed text, wrong-language voice | set the language, pick a voice for that language |
| Voice sounds "different" between chunks | different speed or different voices | one voice per character, Normal speed in all parts |
| The line sounds theatrical | three or more directions per line | keep one or two, leave the rest plain |
| "Voice not found" error | typo in the voice ID or the voice isn't public | copy the ID or fish.audio/m/… link again |
| Wrong word stress | rare word, surname, homograph | spell the word as it sounds or use a synonym |
What to do with the finished voiceover
The file downloads as MP3 or WAV and stays in My Generations. From there you can:
- Make a photo talk — upload a portrait and the audio to OmniHuman 2 or Kling Avatar v2 to get a talking video. Step by step — in the talking photo guide.
- Put it on a video as a voiceover; dubbing an existing video is covered in how to dub a video with AI.
- Build a chain on the canvas: text → voiceover → talking video in one project.
What's next
Start with a short line of 2–3 sentences: pick a voice, add one bracket direction and run 2–3 takes on Drama 3 — in a minute you'll hear how the model plays your text. Then apply the formula to the whole script. The price is shown on the model card before you run, and tokens are bought without a subscription.
Don't want to fiddle with prompts?
Ready-made scenes: upload your photo and get the result in one click.
Models in this article
Frequently asked questions
How do I use Fish Audio text to speech?
Open Fish Audio in the IIshki studio, paste up to 5,000 characters of text, pick a voice from the library, keep the Drama 3 engine and press Create. You get an MP3 or WAV file to download, and if your Telegram is linked, the file is sent to the bot as well.
How do I make Fish Audio sound emotional?
Write a direction in square brackets before the line it applies to: [whispering], [sarcastic], [holding back tears]. Sounds and pauses work the same way — [laughs], [sigh], [pause] — placed exactly where they should be heard. The Drama 3 engine follows directions best.
What is the difference between Drama 3 and S2.1 Pro?
Drama 3 is the most expressive Fish Audio engine: it acts a line out from your directions, can switch delivery mid-sentence, and suits ads, characters and dialogue. S2.1 Pro is faster and cheaper and also understands emotion tags, but reads in a calmer way — use it for long, even narration and drafts.
Can I clone my own voice in Fish Audio on IIshki?
No, the studio has no voice cloning. You get voices from the open Fish Audio library, with filters by language, gender and style. To use a specific public voice from the Fish Audio site, paste its ID or a fish.audio/m/… link into the search of the voice picker.
Is Fish Audio free?
Signing up, choosing a voice and listening to samples are free; you pay only for the voiceover itself, in tokens. The price depends on text length (billed in chunks of 250 characters) and is shown on the model card before you run it; tokens are bought without a subscription.
Every model in this roundup is available in the IIshki studio at reduced club prices. Club members get tokens for generations every month.
Join the clubMore articles
- AI Photo Prompts: 30 Ready-to-Use Prompts for an AI PhotoshootAI photo prompts that work: how a strong photo prompt is built, plus 30 ready examples for photoshoots, portraits, avatars, photo editing and products.
- Nano Banana Pro reviews: what it actually doesNano Banana Pro reviews: what holds up — photorealism, text in images, face consistency, edits from a photo — and where the model still slips. Try it online.
- How to Use GPT Image 2.6: Text, Edits, GIF Stickers and SVGHow to use GPT Image 2.6 by OpenAI: which version to pick, text in images, edits from a reference, animated GIF stickers and SVG vectors, plus common mistakes.