What makes Fish Audio Drama 3 different
Six things set Drama 3 apart from other voice models, and each one changes how you write a script.
| Drama 3 | What it means for your script |
|---|---|
| Direction in your own words | Describe the feeling and the moment, like [proud but exhausted], instead of picking a preset |
| Changes mid-line | Start cheerful, drop to a whisper three words later. Each direction takes over where you place it |
| The hard emotions | Tears under the words, anger held down, a laugh breaking a straight face |
| Word-level control | Paired tags whisper, slow or stress only the words inside them |
| Hard words intact | Codes, dates and place names read right while the voice emotes: 93.2%, the top score at launch |
| Whole scenes | Up to nine voices in one pass, each line answering the last |
Source: Fish Audio's launch post, Artificial Analysis Pronunciation Robustness benchmark.
How to direct a line in Drama 3
A Drama 3 script mixes three kinds of direction, in the same line if you need them.
| Direction | Write it as | What it changes |
|---|---|---|
| Line direction | [warm, steady] before the words | Everything after it, until the next direction |
| Sound or pause | [sigh], [laugh], [breath], [pause] | Adds that sound at that exact point |
| Paired tag | <whisper>And come alone.</whisper> | Only the words inside the tag |
Drama 3 paired tags
Wrap a short phrase in an opening and closing tag. Everything outside it keeps the line's direction.
| Tag | Use it for |
|---|---|
| <whisper> | A secret, a warning, the last words of a threat |
| <emphasis> | The one word the sentence turns on |
| <stress> | Pressing a word harder: YOU were there |
| <soft> | A gentle aside or a reassurance |
| <forceful> | A command or an outburst |
| <slow> | A reveal or a held moment |
| <fast> | A rushed aside or a nervous ramble |
Drama 3 direction tips
Write the note you would give a voice actor in one breath. These habits make the biggest difference.
| Do this | Example |
|---|---|
| Name the feeling and what is under it | [cheerful, but hiding that she is worried] |
| Add the situation or the listener | [quietly, so the kids next door do not hear] |
| A new direction for each mood change | [calm] … [suddenly sharp] … |
| Put a sound after its cause | That's the third time. [sigh] Fine. |
| Match the voice to the role | A soft, breathy voice struggles with a shouted command |
| Render two or three takes | Drama 3 is in preview and takes vary. Keep the best one |
Drama 3 in regional languages
Drama 3 speaks 83 languages. Each sample uses a Fish Audio voice recorded in that region, with the line and the direction both written in the local language. The last card shows the other trick: a direction in one language over a line in another.
Multi-voice scenes in Drama 3
Switch to Dialogue when more than one character speaks: up to nine voices and 10,000 characters per scene, performed in one pass so reactions carry from line to line. Direct each line as you would a single voice, and cast voices that sound clearly different.
Drama 3 vs S2.1 Pro: what changed
S2.1 Pro is Fish Audio's previous model. Both take square-bracket directions, so S2.1 Pro scripts work on Drama 3 as they are. Fish Audio says Drama 3 plays them with more feeling, follows them more closely and reads hard words more accurately.
| Hard words read correctly | S2.1 Pro | Drama 3 |
|---|---|---|
| Overall | 88.1% | 93.2% |
| Names of brands, places and technical terms | 87.3% | 95.9% |
| Words whose reading depends on context | 86.1% | 95.3% |
| Numbers, dates and units | 92.7% | 93.6% |
| Codes, file paths and emails | 79.4% | 86.3% |
Source: Fish Audio's launch post, Artificial Analysis Pronunciation Robustness benchmark.
Drama 3 or Eleven v4 on Morphic
Both are directable voice models on Morphic. Pick by the job, or use both in one project.
| Fish Audio Drama 3 | Eleven v4 | |
|---|---|---|
| Script per speech generation | Up to 10,000 characters | Up to 3,000 characters |
| Dialogue | Up to 9 voices, 10,000 characters | 2 to 10 voices, 2,000 characters |
| Word-level control | Seven paired tags | Tags placed next to the words |
| Languages | 83 | 90+ |
| Voices | Fish Audio voice library | ElevenLabs voice library |
For Eleven v4's tags and pacing, see the Eleven v4 prompt guide.
FAQs
Any plain-language direction in square brackets, such as [nervous, almost a whisper]; sounds and pauses such as [sigh], [laugh], [breath] and [pause]; and seven paired tags around chosen words: whisper, emphasis, stress, soft, forceful, slow and fast.
Write [pause] where the silence should fall. A new direction after it, such as [slow, deadly serious], also changes the pace from there.
Yes. Drama 3 speaks 83 languages, including Hindi, Arabic, Portuguese, Spanish, Mandarin, Filipino and Indonesian, and handles a natural Hindi-English mix. Pick a voice recorded in the region you want, and write the line in that language.
Check the brackets. Directions go in square brackets, and paired tags need a matching closing tag. An unclosed bracket or a mistyped tag can be read as text.
Up to 10,000 characters per generation, about ten minutes of speech, for both Speech and Dialogue. Split longer pieces at natural breaks.
Only with that person's permission: Fish Audio's license for Drama 3 bans cloning a voice without consent. On Morphic, pick a voice from the Fish Audio library.