AI Podcast Voice: Two-Voice Dialogue and Episode Narration Fish Audio

A podcast is a format with a memory: listeners recognize the show by its voices within the first ten seconds of the intro. That is the hard requirement behind every episode — not one good take, but the same voices, week after week, across dialogue, intro and sign-off.

This guide covers how casting, two-voice scripts and episode budgets work here.

Why is a podcast voice harder than audiobook narration?

Audiobooks ask one voice to carry one text. Podcasts ask for three things at once:

• A recognizable pair. The host and the guest are the show's identity; listeners notice a swap instantly.
• Contrast. Two voices in the same register blur together in fast dialogue, especially on a phone speaker.
• Repeatability. Episode 40 has to open like episode 4 — same intro, same sponsor read, same sign-off.

None of those are recording problems here; they are script problems. Settle the voices once, and the work per episode becomes editing, not performing.

How do you cast a show from the library?

The voice library hosts more than 2,000,000 voices. Cast in the order the audience hears them:

1. Host first. The host carries the intro, transitions and sign-off. Pick for endurance, not excitement — you will hear this voice for a full season.
2. Guest against host. Audition the second voice reading beside the host, not alone. Contrast is a property of the pair.
3. Test the handover. Generate a short exchange where the host asks and the guest answers; that turn is where weak casting shows up.
4. Save the shortlist with notes. Write down the register in words so a future episode casts the same way even if you forget the voice's name.

If the library does not have the voice — the host is a real person, or a returning character from an earlier series — cloning is the alternative, and the voice cloning guide covers the recording side.

How does a two-voice episode work?

The engine reads scripts that carry multiple speakers, which turns dialogue into a text problem:

• One script holds both speakers. Mark the speaker changes in the script, assign one voice per speaker, and generate in segments.
• The script is the source of truth. When the guest misreads a line, fix the line in the script and regenerate that segment — the rest of the episode is untouched.
• Direct the delivery inline. Emotion tags such as [excited] or [whispering] travel with the script, so the intro lands the same way every week. The text to speech guide covers how tags and pacing descriptions work.

Generate intro, interview sections and outro as separate segments. Segment files are easier to re-cut in your editor, and one bad line never forces a full re-render.

What if the host voice is a real person?

Cloning builds a reusable voice from a 15 second recording. The practical cases:

• Your own voice, if you want the show in your voice without recording every episode.
• A co-host who records the sample and signs a release naming the show, the scope and the duration.
• A character voice the audience already knows from earlier episodes.

The rights rule is the same everywhere on this site: clone only voices you own or have written permission to use. Keep the release with the project files; the commercial use guide explains what each pack covers.

What does a weekly show cost?

Spoken audio runs at roughly 140 to 150 words per minute, and one credit covers about 1,000 characters:

A weekly 20 minute show is roughly 60 to 65 credits a month — about 6% of the $9 Starter pack. A bilingual feed doubles the generation; even so, a season fits comfortably inside one or two packs. Packs are one-time purchases, credits stay valid for 12 months, and there is no subscription to cancel. Monetized shows and sponsor reads need the $29 Creator pack for commercial usage rights.

Frequently asked questions

Can I generate a two-host conversation in one take?

Yes. The engine reads scripts that carry more than one speaker: mark the speaker changes in the script, assign one voice per speaker, and generate the dialogue. If a single line lands wrong, regenerate that line instead of the whole episode.

How much does a 20 minute episode cost?

Spoken English runs at about 140 to 150 words per minute, so a 20 minute episode is roughly 2,800 words, about 15,000 characters, and therefore about 15 credits. A weekly show lands near 65 credits a month, about 6% of the $9 Starter pack.

Can the host voice be a real person?

Yes, with rights. Cloning builds a reusable voice from a 15 second recording. Use your own voice, or a co-host who signs a release naming the show and the scope. Cloning someone without permission is not allowed.

Does the show voice stay the same across episodes and languages?

Yes. Once a voice is chosen it is reusable for every later episode, and it is not locked to one language: the same voice can read the script in 30+ languages, which is how a show ships localized episodes without re-casting.

Can I monetize a podcast made with these voices?

Commercial usage rights start with the $29 Creator pack, which covers monetized shows, sponsorships and client work. Check the current synthetic media disclosure rules of the platform you publish on before you ship an episode.

Related

• Fish Audio voices - shortlisting a host voice and casting a guest against it.
• Voice cloning - recording a 15 second sample that holds up across a season.
• AI dubbing - shipping the same episodes in more languages with the same voices.