Fish Audio Drama 3 Review: Is It Worth Trying?
By Daniel Walker
Run a script through most TTS tools and you'll get a monotonous, dry read every time. Fish Audio Drama 3 is an exception. You don't just paste text, you brief a voice actor on how each line is supposed to land. It can also process an entire scene with multiple characters.
This review covers what the model does, what it doesn't do and who will benefit from using it. First thing to note: Drama 3 only converts text to speech. It can't change your voice. For that, I've included a section at the end on HitPaw VoicePea.
Part 1: What Is Fish Audio Drama 3?
Drama 3 is Fish Audio's newest TTS model, designed for voice direction and content creation, as the official docs put it. The idea is simple. You write the script, add a few notes on how each line should be played, and the model follows your notes.
A normal TTS tool gives a neutral voice to whatever you enter. Drama 3 feels more like reading a script: the words are the dialogue, and the bracketed notes cover mood and delivery. To get a sigh, a laugh or even a pause, you simply place it where you want it.
It can be accessed through the existing POST /v1/tts request, with drama-3 set in the model header. The same directions work in the web app as well, so you can try it without writing any code. There's one caveat. If the header is missing or misspelled, you won't be notified of an error. The request actually runs on the s2.1-pro model, and it's easy to imagine you're hearing Drama 3 when you're not. It's mainly used for narration, character speech in films, video voiceovers, games and stories.
Part 2: Key Features of Fish Audio Drama 3
Expressive and Emotional Voice Generation
Directions must be put in square brackets, before the words which they should affect. Talk about them as you would for an actor, the emotion, what they want, the speed they speak at, who they speak with. The docs are accompanied by the line "I'm fine. Really. Go on without me. Change the bracket to an angry one or happy one and the same words will sound completely different.
They don't have to be in English. A Chinese note is written on an English line in one example of which there is a record. Therefore, if you have a way of communicating a performance, in any language, then you can ask for a performance.
Multi-Speaker Dialogue and Long-Form Audio
A single request can lead to a complete discussion. Voice IDs are inserted into reference_id and each turn is preceded by a speaker tag, such as <|speaker:0|> or <|speaker:1|>. The demo from Fish Audio is a brief space ship interior scene, featuring a calm captain and a panicked engineer, and a narrator.
The docs don't state any length limit. Animation scripts, drama podcasts and story projects may work better when working on one scene at a time. I would start chapter by chapter and listen for consistency of voices throughout each chapter for a book of this length.
Voice Direction and Script Editing
Brackets are just a part of it. Sounds and pauses can be anywhere they would be deployed naturally, e.g. a shaky breath, a mini-laugh. Only these tags (whisper, emphasis, soft, fast, slow, stress, and forceful) are supported and only cause the words in between them to be modified. Language tags can make a phrase French, Spanish, German, Japanese, Chinese, Korean or Italian, though the docs dub this an experimental feature. Phoneme tags allow you to select the pronunciation and stress, for names and technical terms, these are based on the CMU Arpabet for English.
Everything is plain text and an easy fix for a line is to rewrite the parentheses and regenerate. There is no need to use any audio editing software.
Part 3: Fish Audio Drama 3 Review: Voice Quality, Pros and Cons
A word of disclaimer first. The documentation doesn't contain any benchmark scores for Drama 3, and anything I describe from the audio demos comes from the official examples, not my own measurement. Take each "test" I mention as a challenge you can run yourself.
A good way to test it is to choose one short script and run it several times with different directions, such as sad, excited, angry and hesitant. Next, listen for these four things:
- Emotional expression. On a single line, the official demos deliver a quiet, strained voice well. Check whether your more extreme directions sound acted or just louder.
- Pronunciation. That's why phoneme tags exist. Try brand names and jargon early, and pin down whatever comes out wrong.
- Speaker consistency. With more than one speaker in a scene, check whether each one still sounds distinct after numerous turns.
- Instruction following. Does a direction stay on its own line, or does it flow over into the next one?
Pros
- Expressive results from ordinary-language directions
- Control from a whole line down to a single word
- Multi-speaker interaction in a single request
- Sounds, pauses and pronunciation tags for finer shaping
Cons
- Results vary from one generation to another, so you'll often run a line more than once.
- With unusual emotions, it may take a few quick adjustments to get a good result.
- Changing language mid-line is experimental, and how well it works depends on the language and voice.
- Access and usage limits may vary depending on the plan you select.
The docs say billing is $50.00 per million UTF-8 bytes of input text. A launch promotion cuts that to $10.00 per million until November 12, 2026, 09:00 PT. Chinese or Japanese text costs more per sentence than other text, since bytes are billed rather than characters, and Chinese and Japanese characters take up more bytes.
Part 4: What Can You Use Fish Audio Drama 3 For?
- Audiobooks and expressive narration: A narrator can be calm in one paragraph and tense in the next.
- Animated characters and video dialogue: With multi-speaker generation, you can produce a rough, or even final, dialogue track for a short animation or explainer.
- Games, role-play and character voices: If a line changes, it can be re-recorded in seconds.
- Short films and dramatic storytelling: Inline sounds and pauses make it easy to create a breath or a long silence.
- Creative audio projects and multilingual content: Directions work in any language, and with the experimental language tags it can be used for tutors, travel guides and stories that combine languages.
Keep one limit in mind. Drama 3 is a text-to-speech tool. It can't change a live voice or turn your microphone input into someone else.
Part 5: Want to Change Your Voice? Try HitPaw VoicePea AI Voice Changer
If you want to sound different yourself, a text-to-speech model isn't going to help. You need a voice changer. HitPaw VoicePea, formerly known as HitPaw Voice Changer, alters your voice while you talk instead of creating audio from a script.
Key Features of HitPaw VoicePea AI Voice Changer
- Voice Effects: Offers 300+ voice effects, low latency, and a built-in meme soundboard that supports custom sound effects.
- Compatibility: Available on Windows and Mac, supports 16 languages, and offers a free version.
- Gaming Voices: Promotes character voices for games such as Fortnite, GTA, and Minecraft.
- Streaming Support: Allows streamers to change their voices in real time on Twitch, YouTube, and Kick.
- Calling Apps: Supports Zoom, Microsoft Teams, Google Meet, and Discord.
- File Voice Changing: Applies AI voices to existing audio and video files in 50+ formats.
- AI Cover Songs: Allows users to create AI cover songs.
- Hardware Requirements: According to the user guide, real-time AI voice effects require a reasonably capable graphics card, with Nvidia 4 GB and AMD 6 GB-class devices recommended.
How to Use HitPaw VoicePea AI Voice Changer
The official guide reduces setup to a few steps:
Step 1:Download and install VoicePea for Windows or Mac from the product page. The free version is enough to test it.
Step 2:Open Settings and pick your real microphone and headphones as input and output. If you can't hear anything, click "Detect"; the microphone and speaker icons should turn green.
Step 3:Choose a voice or sound effect and click to apply it. Hover over the cartoon avatar to adjust volume, pitch and timbre.
Step 4:Switch on "Voice Changer" and "Hear myself," then talk to check the result. In a noisy room, turn on noise reduction, but keep it low, since a high setting can filter out parts of your voice.
Step 5:In the app you'll be using, such as Discord or Steam, open the voice settings and set the input device to "HitPaw Virtual Audio".
Use headphones throughout. Speakers leak into the microphone and create a feedback loop. If no one can hear you, the most likely issue is that the wrong input device is selected.
Fish Audio Drama 3 vs. VoicePea: Which One Should You Choose?
The only difference is where each one begins. Drama 3 starts from a written script and creates expressive speech. VoicePea starts from your voice and morphs it, live or on a file.
Use Drama 3 for scripted narration, multi-character dialogue or any task that requires directing each line. If you want to sound like another person in a game, stream or call, or even alter an existing recording, use VoicePea. You can also mix them: generate the dialogue in Drama 3 and use VoicePea for the parts you perform yourself.
FAQs about Fish Audio Drama 3
It's Fish Audio's text-to-speech model for directed performances.
According to the docs, it costs $50.00 per million UTF-8 bytes, with $10.00 during the launch period until November 12, 2026. That page doesn't mention a free tier, so please check Fish Audio's pricing page for current plans and credits.
Yes. Include several voice IDs in reference_id and add a speaker tag to each turn. One request can produce a complete scene.
Fish Audio documents voice cloning as a stand-alone feature, and the Drama 3 page mentions using a cloned voice from a bilingual speaker for language switching. Make sure you only clone voices you have permission to use.
Conclusion
What impresses me most about Drama 3 is the control it gives you over delivery. Bracketed directions, inline sounds, word-level tags and multi-speaker scenes from a single request make it suitable for narration, animation dialogue and character-driven stories. However, expect a bit of learning by doing. Results differ from one generation to the next, and what you can do depends on your plan.
For expressive AI narration, it's worth a try. For those who want to alter their own voice, HitPaw VoicePea is a convenient option.
Leave a Comment
Create your review for HitPaw articles
