Most podcast tooling assumes one show means one language. If you want a Spanish version you make a second show, find second hosts, and maintain two production pipelines that drift apart within a month.
Podcast Creator Studio treats language as a property of the episode, not of the show. One podcast definition — one cast, one topic, one schedule, one set of sources — produces a complete episode in each language you enable, every time it runs.
What actually gets produced
Enable two languages and a single run gives you two of everything:
- A script per language
- An audio file per language —
podcast_English.mp3,podcast_Russian.mp3 - A video per language, if you have video generation switched on
- Episode metadata per language
These are separate finished artefacts, not one file with subtitles bolted on. Each can be shared, emailed or published on its own.
Generated, not machine-translated
There is an important distinction in how each language is produced, and it is configurable per language.
Generate writes the episode natively in that language. The script is composed in Spanish, from the sources, by a model working in Spanish. Idiom, sentence rhythm and the way a host would naturally phrase a point all come out right, because nothing is being carried across from an English original.
Translate takes the script written in your base language and renders it into the target. This is faster and keeps the two versions structurally identical — the same beats in the same order — which is what you want when the versions need to correspond closely.
The rule of thumb: use generate when the audience for each language is a separate audience, and translate when you need the versions to line up.
Voices per language, not one voice stretched across all of them
A voice that sounds natural in English does not sound natural reading Russian. So voices are assigned per character per language: your host ALEX can have one voice for the English episodes and a different, native-sounding one for the Spanish.
There is a free tier of voices covering a wide range of languages, and if you bring an ElevenLabs key you can use their voices — including cloned ones — instead. Voice assignment falls back sensibly: a character with no explicit voice for a given language inherits a suitable default for that language rather than being forced through the wrong one.
One detail worth knowing, because it is the kind of thing that goes wrong elsewhere: the studio checks that a male character does not get assigned a female voice, or vice versa, when falling back to a language default.
Setting it up
- Open Settings → General → Target Languages and enable the languages you want available across the account.
- On the podcast's own Settings tab, choose which of those this particular show generates. A podcast inherits the general list unless you override it.
- On the Cast tab, assign each character a voice for each language. Anything you leave blank inherits a default.
- Generate. Every enabled language is produced in the same run.
Note that your base language must be set to generate rather than translate — there has to be an original for anything to be translated from.
Publishing multilingual episodes
Each language's audio file is separately shareable. Under the Share tab you will find a link for every language the run produced, alongside the RSS feed and cover art. Sharing a bilingual show to Slack or WhatsApp carries both language versions rather than silently dropping one.
If you publish an RSS feed, the feed has a declared language, and the episode enclosure it serves is the one matching that language. Running a separate feed per language is the conventional approach and the one podcast directories expect.
Where this earns its keep
- Organisations with a mixed-language workforce. The same internal briefing, in the languages people actually think in.
- Education. A lecture series produced simultaneously for several cohorts.
- Reaching a second audience without doubling the work. The marginal cost of the second language is a generation run, not a second production team.
The point is not that translation is hard — it is that maintaining two shows in parallel is hard. Making language an attribute of the episode removes the parallel maintenance entirely.