A scripted audio drama asks a listener to hold an entire cast in their head with no faces attached. Character identity lives in the voice: its pitch, its tempo, how it clips a consonant when the character is angry, how it softens when the character is lying. That is what makes audio drama localization a different discipline from translating a script. The words can be translated in an afternoon. The cast has to be rebuilt.

In picture dubbing, a voice that drifts slightly from the original performance can be carried by the image. The face, the wardrobe and the blocking keep identity attached to the character. In audio fiction there is nothing to carry it. A listener who loses track of who is speaking does not miss a line, they lose the scene, and they spend the next thirty seconds reconstructing the last one instead of following the plot.

This guide works through the decisions that determine whether a translated season holds together: casting and freezing a target-language voice map, making characters distinguishable by ear alone, directing emotional performance when the voice is generated, protecting sound design and music while dialogue is replaced, adapting scripts where wordplay or dialect carries the plot, holding continuity across episodes, catching voice drift in review, and handling release timing, rights and credits.

What audio drama localization has to solve that dubbing for picture does not

There is no visual anchor

Picture dubbing tolerates a certain amount of vocal mismatch because the eye keeps the ledger. Remove the picture and the voice is the entire identity. Once two characters have been confused, the listener starts auditing the production rather than following the story, and the confusion compounds across scenes because they are now tracking the production as well as the plot.

Attention is the only budget

Distinctness, dialogue intelligibility, music level, pacing and the number of named characters introduced in a scene compete for the same working memory. They are not independent choices. Translated dialogue also changes length: scripts frequently expand when moving between languages, and a longer read either speeds up the performance, which flattens it, or lengthens the scene, which changes the rhythm of the episode. Both outcomes are decisions, and neither should be made accidentally in a mixing session.

The mix is the set

In audio fiction, sound design is production design. Room tone says kitchen, a specific reverb says stairwell, a music cue marks the act break. A translated edition has to re-earn all of that while the dialogue on top is replaced. The workflow most teams borrow from dubbing for picture is a reasonable starting point, but it assumes clean stems and a picture edit to conform to, and an audio drama often arrives with neither.

Building the target-language voice map

Cast from the character bible, not from line count

Build a table before casting anything, one row per recurring character, with the fields that actually constrain a voice:

  • Function in the ensemble, and who they are in conflict with.
  • Approximate age, vocal weight (light, medium, heavy), habitual register.
  • Baseline tempo and articulation habit, such as clipped or drawling.
  • Regional, class or generational accent, and whether it carries plot weight.
  • The two or three scenes that define how this character should sound at their extremes.

Casting from the script alone invites collisions, because two characters who never share a scene in episode two may share the entire climax in episode nine, and by then the casting decision is expensive to revisit.

Fix bands, not descriptions

Give each character a band: a range of fundamental pitch, a tempo range, and a rule about articulation. Three constraints do most of the work.

  • No two characters who share a scene may overlap in both pitch band and tempo band. Overlap in one dimension is survivable; overlap in both is where attribution collapses.
  • The two leads should be separated in whichever dimension carries the story's central contrast. If the drama is about calm against volatility, separate tempo rather than pitch.
  • Characters who appear once per episode can sit closer together than series regulars, because the listener has less to hold.

Lock the map before episode one airs

Voice changes mid-season are the most common failure in translated audio fiction. Once the pilot passes the attribution test, freeze the map and record reference lines for every recurring character. A replacement in episode nine then has a target to match instead of a memory to argue with.

Making characters distinguishable by voice alone

The blind-listen test

Take a scene with three or more speakers, present a version with the names stripped from the dialogue, and ask listeners who have never heard the show to attribute each line. If two listeners disagree about who said a line, the scene is not finished, regardless of how good the translation is. Run it on the pilot before committing to a season, and again on any episode with a new character or a large ensemble scene.

Use diarization as a continuity instrument

Speaker diarization on the source season produces a speaker-turn timeline: who speaks, when, and for how long. Run the same analysis on the translated episode and compare the two. If the turn boundaries land in roughly the same places, the adaptation has preserved the conversational shape of the scene. Where a turn moves, either the translation deliberately reordered dialogue or a line was assigned to the wrong voice. Crowd scenes, overlapping dialogue and whispered exchanges are the usual sources of error. The speaker labels and timings are available programmatically if you want the comparison automated, as described in the platform documentation.

Secondary cues carry the rest

Beyond pitch and tempo: forms of address, name-drops, a signature laugh, a repeated interjection, a consistent distance from the microphone, and consistent placement in the stereo field. These are cheap and they do a surprising amount of identity work in scenes where the dialogue is deliberately terse.

Directing emotional performance in generated speech

Turn notes into parameters

"Furious" is not direction. Direction describes the acoustic result: where the emphasis lands, where the pauses fall, whether volume rises or drops, and how the line starts relative to the character's baseline. A usable note reads: begin slower than baseline, hold the pause before the name, drop volume on the final clause. That note works for a human actor and it is renderable when the performance is produced through speech generation, where tempo, pause placement and emphasis can be set per line rather than per take.

Regenerate rather than re-book

A synthetic pipeline lets you revisit a single line in episode six without assembling a cast. Use it, but keep the change small. A regenerated line changes duration, and duration changes how the line sits against a music cue or an overlapping effect. After any regeneration, re-run the continuity and mix checks on that scene.

Cloning and consistency

Where a performer authorizes it, voice cloning holds a timbre steady across a season and across languages, which solves the recasting problem structurally: the voice does not change because a schedule did. The trade-offs are contractual and procedural. You need explicit authorization with a defined term and scope, and a process for re-synthesizing a line when the script changes rather than calling the actor back. Treat the voice model as a cast member with a contract.

Keeping sound design, music and effects intact

Start from stems

The single decision that determines how hard a translated season will be is whether the original exists as separate dialogue, music and effects stems. With stems, dialogue replacement is a mixing job. From a single stereo master, it is a rebuild, because extracting dialogue also removes the room tone and any effects sharing its frequency range. If a translated edition is even possible in the future, request stems at the mix stage; it costs almost nothing then and a great deal later.

Decide what survives

  • Music beds, designed transitions and ambience loops usually survive cleanly when the licenses allow.
  • Dialogue-adjacent sounds, such as a glass set down under a line or clothing movement, often need re-creation from a library.
  • Room tone beneath dialogue has to be replaced, or new dialogue sits in an unnatural silence.
  • Cues timed to the source language's rhythm may need small shifts, particularly where a line lands on a musical accent.

Mix for words, not atmosphere

A translated mix usually needs a more dialogue-forward balance than the original, because the listener is hearing unfamiliar names and places for the first time and has less spare attention. Check the result on headphones, a phone speaker, and in a car. Those are the three places audio drama is actually heard, and intelligibility fails differently in each.

Adapting wordplay, accents and dialect as plot devices

When the joke is the language

Puns, homophones and idioms that carry plot weight have three honest resolutions: replace them with a target-language joke that delivers the same character beat, keep the literal meaning and lose the humor, or cut them and compensate elsewhere in the scene. There is no fourth option where the joke survives translation intact. Decide per instance and record the decision in the continuity ledger, so episode four does not resolve a running gag differently from episode one. A running gag that changes shape between episodes reads as a mistake even when each individual choice was defensible.

Dialect as a reveal

If an accent is the twist, removing it kills the twist. A character whose regional accent marks them as an outsider, or whose code-switching marks a change in who they are addressing, carries information in the voice itself. The target version needs a variant that carries a comparable social signal for the target audience, which means casting for it deliberately rather than hoping a general accent will cover it. Where no equivalent exists, the adaptation has to move the signal into the dialogue with a line that does the work the accent was doing, and accept that the scene is now slightly more explicit than the original.

Set the accent policy before casting

Choose one: cast speakers of the dialect, cast performers who can execute it credibly, or drop the dialect and compensate in writing. The failure mode is inconsistency, with one character's accent rendered three different ways across a season by three different performers who each received different instructions.

Episode continuity and the listening-review protocol for audio drama localization

The continuity ledger

One document, shared with every translator, adapter, director and mixer on the season, containing:

  • Character names and their approved pronunciation, with a phonetic spelling and a reference recording.
  • Invented places, organizations, technologies and titles.
  • Recurring phrases and catchphrases, with the approved target rendering and the episodes where they appear.
  • Forms of address, and how they shift when relationships change.
  • Units, currency and dates, and whether they are converted or left alone.

Freeze pronunciation decisions

Names are the most common source of drift. Decide the stress pattern for every proper noun early, record it, and treat it as fixed. A character whose name is stressed differently in episode seven than in episode one sounds like a different character to a listener who has heard all six previous episodes.

The six-pass listening review

  1. Comprehension pass. Listen to the finished episode with the script closed and note timestamps where you lose the thread of who is doing what.
  2. Attribution pass. In every scene with three or more speakers, confirm you can name the speaker of each line without looking at the script.
  3. Continuity pass. Check the episode against the ledger: catchphrases, pronunciations, address forms, numbers.
  4. Mix pass. Headphones, phone speaker, car. Test dialogue intelligibility against music in each.
  5. Voice-drift pass. Play a reference line from episode one and the equivalent line from the current episode back to back. Any change in pitch, tempo or articulation beyond the character's band is drift.
  6. Cross-episode pass. Play the last two minutes of the previous episode and the first two minutes of the current one without a break, which is how a returning listener hears them.

Run the full sequence on the pilot, then sample on later episodes, treating any defect found in a sample as a trigger to run the full sequence on neighboring episodes.

Release strategy, rights and credits for a translated season

Simultaneous or staggered

Simultaneous release across languages buys one press cycle and a single marketing moment. It also requires every language to be finished and reviewed before episode one ships, which means the voice map is frozen before any audience has heard it and any systematic problem multiplies by the number of languages. A staggered release lets you pilot one language, run the review protocol, correct the map and the ledger, then batch the remainder against a tested template. The costs are a split press cycle and spoilers leaking from earlier languages into later ones. For a first translated season, pilot-then-batch is usually the better trade. For a returning series with a proven map and a locked ledger, simultaneous becomes reasonable.

The rights matrix

Rights for a translated audio season are several clearances, not one:

  • Script rights, since a translation is a derivative work that the original writer's agreement typically has to name.
  • Performance rights for any reused recordings from the original cast.
  • Voice-model rights for any cloned voice, with an explicit term and a scope covering languages and territories.
  • Music and effects licenses, which are often territory-limited, so a track cleared for one country may not be cleared for a translated edition released in another.

Settle these before mixing, not before release. If you need to model cost per language before committing, the current options are listed on the pricing page.

Credits and billing

A translated edition needs its own credit block: original writer, translator, adapter, voice director, cast, sound designer, mixer. Translation credits are routinely dropped, which is both inaccurate and a practical problem, because uncredited translators cannot point to the work when the next season is staffed. Where a rights position is unclear, or a performer's authorization needs to extend to more languages than originally agreed, raise the question with the team early enough that the answer can still change the plan.

Frequently asked questions

Does voice cloning require the original performer's authorization?

Yes. Cloning is used to preserve a speaker's voice across languages, and it depends on explicit authorization with a defined term and scope. Treat the agreement as a cast contract rather than a file transfer.

Can a translated season reuse the original music and effects?

Often, if the original was mixed with separate stems and the licenses cover the release territory. If you only have a finished stereo master, plan for reconstruction of room tone and dialogue-adjacent effects.

What do you fix first when listeners say they cannot follow who is speaking?

Run the attribution pass on the specific scene. The cause is usually overlapping pitch and tempo bands between two characters who share the scene, or a mix where both voices sit at the same distance and position.

How do you keep a catchphrase consistent across a whole season?

Freeze it in the continuity ledger with the approved target rendering and the episode list, and include the line in the voice-drift pass for every later episode. Re-deciding a catchphrase in episode eight is the most common continuity break.

Is a simultaneous multi-language launch a good idea?

For a first translated season, rarely. Pilot one language, review it fully, correct the voice map, then batch the rest. Simultaneous makes sense once the map and ledger have already survived an audience.

Can generated dialogue be mixed with original cast recordings?

Yes, if you have stems and the performance sits in the same acoustic space. Match room tone, distance and level first, then judge the blend, because a technically correct line in the wrong space sounds worse than a slightly imperfect line in the right one.

Conclusion

Treat the voice map as the deliverable, not the translated script. A season that reads beautifully and cannot be followed by ear has failed at the only job audio fiction has. Before committing to a full translated season, produce one pilot episode in one target language, run the six-pass review on it, and use what that review finds to finalize the pitch and tempo bands, the ledger and the pronunciation list. That single episode costs a fraction of a season and tells you whether the cast you have cast can actually carry the story without pictures.

Then decide the release shape with real information in hand. If the pilot holds, batch the remaining languages against the frozen map and ship staggered, with the ledger shared across every language team from the first session. If the pilot reveals collisions, the fix is nearly always casting and band assignment, not translation, and it is far cheaper to fix at episode one.

Finally, resolve rights and credits before the first mix session rather than after the last one. Translation rights, performance rights, voice-model authorization and territory-limited music licenses each have a lead time, and each can invalidate work that has already been recorded. Get the credit block drafted at the same time, with the translator and adapter named, so the people who built the translated season are visible when the next one is commissioned.