How do I play several characters myself in an audio drama?

This page may contain affiliate links from which Blur Studio earns commissions.

This is one of the more creatively satisfying uses of voice-changing tools, and it genuinely works, with a caveat: you still have to act each character differently. Voice Changer changes what you sound like, not how expressively you perform, so the groundwork of a convincing audio drama is still on you as the performer. If you would rather not perform every part, voicing a script with several characters uses a different route.

The actual workflow

1. Act each character's lines separately, in your own voice, changing your real delivery for each one: pacing, tone, energy, accent if you can manage it. This is the part that makes the final result convincing or flat, and no voice conversion fixes a performance that's acted the same way for every character.
2. Pick a distinct target voice per character from ElevenLabs' Voice Changer, drawing from its 10,000+ Voice Library options, designed voices, or clones you have rights to. Choose voices that are genuinely different from each other, not just different names, since two similar-sounding target voices will blur together for a listener even with different underlying performances.
3. Convert each character's recording separately, matching it to that character's target voice.
4. Join the converted clips together in your audio editor, adding sound design, music or ambience as needed.

One real production limit to plan your recording sessions around: each conversion caps at roughly 5 minutes on most plans. For a scene longer than that, record and convert that character's lines in multiple segments rather than trying to push through one long file, and the uploaded file itself also needs to stay under 50MB. If one line comes out wrong later, see fixing one wrong sentence without a full re-record.

A worked example: a three-character scene

Say your script has a tense three-way exchange, a detective, a nervous witness and a skeptical partner, all talking over a short scene. Here's the actual sequence to pull it off convincingly with one performer:

1. Decide each character's performance baseline before you record anything. The detective is slow, flat, deliberate. The witness is fast, breathy, cutting herself off mid-sentence. The partner is dry, slightly bored, clipped. Write these one-line performance notes down; they're what you'll actually be acting, not the voice identity.
2. Record the detective's lines first, straight through the whole scene, at his pace, ignoring the other characters' timing for now. Don't try to leave realistic gaps for where the others will speak; you'll handle timing in the edit.
3. Record the witness's lines next, performing her distinct energy, faster, more anxious, with real pauses where she'd actually hesitate. This is where the performance gap between characters actually gets created, not in the voice conversion step.
4. Record the partner's lines last, dry and clipped, deliberately underplaying compared to the other two.
5. Convert each character's full take through Voice Changer to its own target voice (three genuinely distinct voices from the Voice Library, chosen so their natural pitch and pacing don't sit too close together), then assemble the converted clips in your editor against a rough timeline, adjusting gaps so the exchange actually flows like three people in a room.

The scene works because the detective, witness and partner were three different performances before any voice conversion touched them. The voices make them sound like different people; the acting is what makes them feel like different people.

What actually makes this convincing, or not

The single biggest factor in whether a one-person cast sounds believable isn't the voice choice, it's how differently you actually perform each character before conversion ever happens. A flat, similar delivery across characters, converted to different voices, still sounds like one actor doing different voices badly, just with different pitch on top. Lean into real differences in pacing, pausing and emotional register between characters when you record, and the conversion step will carry that performance through convincingly.

Second, don't rely on the voice choice alone to distinguish characters in a scene with rapid back-and-forth dialogue. If two characters' target voices sit close in pitch or tone, listeners can lose track of who's speaking during fast exchanges, even with a clearly different voice technically applied. Picking contrasting voices up front, not just different ones, saves you from needing to fix this in editing.

The settings worth knowing about

Once you've converted a take, four sliders let you fine-tune the result: Speed, Tone, Pacing and Style Exaggeration. These are for small shaping, not for fixing a flat performance; cranking Style Exaggeration on a character who was acted the same way as everyone else makes that voice sound more stylized, not more like a genuinely different person. Where these sliders earn their keep is consistency: write down the exact settings you land on for each character the first time, so episode two, scene four, or six months later when you need one more line from that character, you're reapplying known-good numbers instead of trying to remember or re-guess them. Output comes back as a clean WAV file, which is what you want feeding into a proper audio editor rather than a compressed format that'll lose quality again when you process it further. For the single-voice version of this workflow, see changing the sound of your voice while keeping your own delivery.

Roughly what this costs for a short drama

Voice Changer runs on credits, and a rough rule of thumb is about 1,000 credits per converted minute. A 20-minute episode split across three characters, say 7 minutes of dialogue per character once you account for all three parts, costs roughly 7,000 credits per character, 21,000 credits for the whole episode. On a Starter plan's 30,000 monthly credits, that's comfortably one full episode a month with room to spare for a retake or two; a Creator plan's larger allowance covers several episodes a month. Compare that to even a modest budget for three voice actors' session fees for the same 20 minutes, and the economics of doing it yourself with voice conversion are obvious for anything beyond a one-off project. Making different character voices for a course or game covers the non-drama version.

Casting the same characters in another language

If you're producing a version of the same drama in a second language, Arabic alongside an English release, say, the cleanest approach is to re-record your own performance in that language rather than trying to dub the English conversion afterward. Your delivery, pacing and emotional beats in Arabic won't be identical to your English take line for line, since language itself changes rhythm, so performing fresh in each language and converting each separately keeps every character's performance genuine in that language rather than feeling like a translated-and-pasted-on voice. Keep the same target voice per character across both language versions; that's what makes a listener who hears both recognize the same "cast" even though the actual language changed. One practical note for Arabic specifically: the model doesn't expose a dialect picker, so you'll get a standard Arabic register rather than a specific regional accent, which is fine for most content but worth knowing before you promise a particular dialect to an audience that would notice the difference.

When two characters still blur together after conversion

If you've picked what look like two different target voices and a scene between those characters still confuses a listener, the actual problem is usually that the two voices sit too close together in pitch and pacing, even though they're technically different voices with different names. The fix isn't more editing in post; it's going back and picking voices that contrast more sharply before you record, ideally testing a short exchange between just those two characters early rather than discovering the problem after converting a full scene. A quick trick: if you can't tell the two target voices apart from their names and a short preview alone, before any of your own performance is layered on top, a listener won't reliably tell them apart once your performance is added either.

How I'd approach this

For a short audio drama or a one-off project, doing this yourself scene by scene is entirely workable, and it's genuinely one of the more fun small production workflows voice AI has made possible for a single person. If you're producing a longer series with a recurring cast, I'd set up a small reference library mapping each character to its target voice and exact slider settings once, so you (or whoever's editing) aren't re-deciding those choices for every new episode, just applying what's already locked in. For a production with enough episodes that this reference library becomes its own small management problem, a lightweight internal tool that stores each character's voice ID and settings, and lets whoever's producing an episode pull them up instead of hunting through old notes, is a genuinely useful thing to build once the manual version starts creaking.