Fish Audio Explained: How to Make Your First Voice-Over
What Fish Audio makes, which voice model to use, what the free plan allows, and how to turn a script into a voice-over.
Published

Fish Audio turns written scripts into speech and can clone a voice from a recording. The free web plan includes up to 7 minutes of speech a month for personal use, with no credit card required. For a client video, the first thing to settle is whose voice you have permission to use.
How Fish Audio text to speech works
Think of a voice-over as a casting decision followed by a recording session. In Fish Audio, the voice you choose plays the part of the narrator. The script supplies the words. A performance direction tells the speech model how to read them. If a line calls for a whisper, you can change the direction while keeping the same narrator and script.
Hanabi AI Inc. provides the service. Its official domain is fish.audio. If you land on fishvoice.app, you’re on a different domain. Start at the official site for the account, prices and terms discussed here.
The voice library includes official and community voices. Fish Audio advertises more than two million voices, but you can make a casting decision with one sentence. Pick a line that contains the name of your product and a change in mood. Keep the wording fixed when comparing narrators so you can hear which choice made the difference.

When to use the Fish Audio voice changer
If you have a recorded performance and want to change the speaking voice, the Voice Changer takes audio as its input. If you have a written scene with several speakers, Story Studio lets you assemble their lines in a long-form project. Both are in the web app guide.
How to start with the Fish Audio AI voice generator
For a first personal project, use a library voice and a short passage. You can review the library and the public demo before creating an account. Fish also offers iOS and Android apps for generating speech on a phone.
Pick a passage you can judge
Take a line from the recording you want to make. A museum guide might use “The last ferry leaves at dusk.” Decide which word the listener needs to remember.
Choose a voice and open Text to Speech
Browse the library and choose Use on a voice you want to try. When prompted, the account creation page asks for your email. In Text to Speech, paste the passage and select your voice and model.
Direct one line and listen back
For scripted delivery, Fish Audio Drama 3 accepts directions in square brackets. Try putting
[quietly]before the ferry announcement. Generate the line and listen for the emphasis on “last” and the pronunciation of “dusk.” Change one thing at a time.Save the take
Open Generation History to replay and download the recording. Before putting the take in a paid tour or client video, check the plan terms and your rights to the voice.
What Fish Audio voice cloning requires
Cloning uses a recording as the reference for how a speaker sounds when saying new words. The recording and the speaker can have different rights holders. Owning the file from a podcast interview, for example, doesn’t tell you whether the guest agreed to narrate your next episode.
Fish Audio’s Terms of Use require the content owner’s prior consent when you upload or use material you don’t own. Its Professional Voice Clone guide also requires that the voice be yours or that the speaker consent. That workflow includes verification and needs a paid plan with an available professional voice slot. Ask the speaker about the specific use, such as narration for a paid ad, before uploading a sample.
What the Fish Audio free plan includes
Free Tier includes 8,000 credits each month, up to 500 characters per generation and three public voice slots. For a private rehearsal, that gives you room to revise a few short lines. For a sponsored video, the pricing FAQ restricts free output to personal, noncommercial projects and describes paid commercial use for verified voices you own.
Paid creator plans start at $15 on regular monthly billing. On November 22, 2026, the USD pricing page also showed first-year annual offers of $66 for Plus and $450 for Pro, both billed upfront.
The last two rows belong to the separate voice-agent plans. A team generating prerecorded narration can use the creator rows. A team building telephone conversations needs the call rate and the number of simultaneous calls its agent plan allows.
Which Fish Audio model fits the job?
For scripted narration, start with Drama 3. You can place a direction where a sentence changes mood. Fish documents its bracket syntax and gives side-by-side readings of the same line in the Drama 3 release guide.
For a conversation, consider Fish Audio Live 3. It supports English and listens while it speaks, a capability called full duplex. If a caller interrupts a booking request to change the day, the system needs to make room for the correction. That’s a different audition from reading a paragraph aloud. Fish illustrates this behavior in its Live 3 announcement.

When you need a Fish Audio API key
Get a key under API Keys in the web app when your own product needs to request audio. A creator subscription is optional, and the API has separate billing. In the text-to-speech quickstart, the app sends “Hello from Fish Audio!” to POST /v1/tts and saves the returned audio as out.mp3. To keep the same narrator across different messages, send the voice’s identifier in reference_id.
For a prototype, the developer model catalog lists s2.1-pro-free under fair-use limits. It uses the same model as s2.1-pro, the recommended production option in the S2 family. Fish offers no time-to-first-audio guarantee or data processing agreement for that free model. Those details matter if a customer is waiting on the line or your buyer requires a contract covering their data. Compare speech APIs for live voice agents against those requirements.
Older examples may name s2-pro, the previous S2 model, or s1, which Fish schedules for retirement on December 31, 2026. For the opposite task, turning a recorded interview into text, Fish recommends transcribe-1-pro.
Where to go next
ElevenLabs is the established choice to compare when editing takes matters as much as generating them. Its Studio lets an editor trim audio and regenerate individual lines. Bring the same paragraph to both editors and try replacing a single sentence after the script has been approved.
Murf is worth a look for studio voice-over if your editor needs to adjust pitch and speed to fit a training video. For a broader shortlist, see the guide to AI voice generators or the ElevenLabs alternatives by workflow.
If your shortlist is already down to two, the Fish Audio vs ElevenLabs comparison covers the choice for narration, cloning and live calls.
- Published
- Category
- AI










