The engine choice shapes how quickly you can refine a script, how a performance lands, and which languages your project can serve. Treat it as a production decision, not a setting to guess at after the edit is finished.
Vois keeps that decision inside the same project as your script, voice assignments, timeline, and export. You can audition a direction with the fast engine, move the approved English performance to the expressive engine, or use multilingual coverage without rebuilding the project around a new service.
Use this guide to choose a starting point, then listen against the actual script and footage. The useful outcome is a repeatable workflow that gives each part of the job the right balance of speed, delivery, and language coverage.
The fast engine: speed for iteration
Speed: 3-6x real-time (6x with GPU acceleration on Apple Silicon) Languages: English only Best for: Drafts, rapid iteration, high-volume batch generation
The fast engine does exactly what the name suggests. A 10-minute script generates in about 2 minutes on a modern machine. With GPU acceleration on Apple Silicon Macs, it can be quicker.
Why does speed matter so much? Because the real workflow for voice content isn't "write, generate, done." It's "write, generate, listen, rewrite that awkward sentence, regenerate, adjust the pacing, regenerate again." Iteration is where good audio becomes great audio.
When you're cycling through variations of a script, waiting 10 minutes per attempt kills momentum. The fast engine keeps you in a creative flow. Generate, listen, tweak, generate again. The gap between idea and output stays short enough that you don't lose the thread of what you're trying to achieve.
The trade-off is emotional range. The fast engine produces clean, natural-sounding speech, but it doesn't handle dramatic inflection or paralinguistic sounds the way the expressive engine does. Think of it as a solid, professional read. Not flat, not monotone, but not theatrical either.
Use it when: You're drafting and iterating. You're producing straightforward narration (explainer videos, documentation, tutorials). You're doing batch generation where speed matters more than emotional nuance. You're previewing how a script will sound before committing to a slower, higher-quality generation pass.
The expressive engine: emotion and natural inflection
Speed: 2x real-time Languages: English only Best for: Podcasts, audiobooks, character dialogue, anything that needs to feel alive
This is the engine that makes people do a double-take. The expressive engine generates speech with natural emotional inflection, and it supports emotion tags that give you direct control over how specific lines are delivered.
Drop a [laugh] tag before a sentence and the voice actually laughs. Add [sigh] for a weary delivery. Use [shush] for an intimate, secretive tone. These aren't gimmicks. They're the difference between AI narration that sounds like it's reading and AI narration that sounds like it's performing.
For podcasts, the expressive engine is the obvious choice. Conversational delivery with natural pauses, emphasis on the right words, and the kind of vocal variety that keeps listeners engaged over 30-60 minute episodes.
For audiobooks (especially fiction), it's even more important. Characters need to sound different from the narrator. Emotional scenes need to land. Dialogue needs rhythm and personality. The expressive engine handles all of this.
The speed trade-off is real but manageable. At 2x real-time, a 10-minute script takes about 5 minutes to generate. That's fast enough for production work, just not ideal for rapid-fire iteration. The smart approach: draft with the fast engine, finalize with the expressive engine. More on that workflow below.
Use it when: You're producing final-quality audio for podcasts, audiobooks, or any content where emotional delivery matters. You need character voices that feel distinct and alive. You want control over specific moments using emotion tags.
The multilingual engine: 23 major languages, one workflow
Speed: 1x real-time Languages: 23 major languages (English, Spanish, French, German, Italian, Portuguese, Japanese, Chinese, Korean, Hindi, Arabic, and more) Best for: Non-English content in major languages, localized versions of existing content
If you're producing content in a major non-English language, this engine delivers high-quality pronunciation and intonation patterns specific to each language. It's not English speech with an accent bolted on. Each language model understands the rhythm, stress patterns, and phonetic rules of its target language.
A common use case: you've produced an English podcast or course, and now you want to reach Spanish, French, or Japanese audiences. Feed the translated script to the multilingual engine, pick a voice, and generate. The same workflow, the same project structure, just a different language.
The speed is roughly 1x real-time (a 10-minute script takes about 10 minutes to generate). That's because multilingual models are larger and more complex. Plan your generation time accordingly, especially for long-form content.
Omni: 600+ languages and voice design
Languages: 600+ (every language the multilingual engine covers, plus hundreds more) Best for: Less-common languages, voice design, maximum language coverage
Omni expands coverage to 600+ languages, including many languages beyond the multilingual engine's 23. Use it when the project needs a less-common language, a consistent localization workflow, or a designed voice that fits a defined role.
It also adds voice design. Select a set of supported voice attributes, review the resulting candidate, and save the one that suits the production. Pair that with properly consented voice cloning when a project needs an approved reference voice in its palette.
Use it when: You are localizing for languages outside the legacy multilingual set, need broad language coverage, or need an approved designed voice. Have a qualified reviewer check translated content and pronunciation before export.
The decision guide
Not sure which engine to use? Walk through this:
| Question | Best starting point |
|---|---|
| Are you drafting an English script and expect several revisions? | Start with fast |
| Does an English podcast, audiobook, or character scene need a more performed delivery? | Use expressive |
| Is the project in one of the 23 legacy multilingual languages? | Use multilingual |
| Do you need broader language coverage or Voice Design? | Use Omni |
| Are you producing straightforward English narration in volume? | Start with fast, then listen before export |
| Are you unsure? | Generate a short representative scene in the likely engine, compare it against the edit, then choose |
The choice is reversible. In Vois, your script and voice assignments stay with the project while you change the engine, so you can compare a representative scene before you commit to a full render.
The two-pass workflow: draft fast, finalize expressive
The practical two-pass workflow combines fast iteration with a more expressive final performance.
Pass one: Generate the project with the fast engine. Listen for pacing, unclear sentences, and voice assignments that do not suit the material. Make structural script edits in this pass.
Pass two: Switch to the expressive engine for the approved English scenes and regenerate the final takes. For localization, follow the same review sequence with multilingual or Omni and a reviewer who can verify the language.
Vois makes the handoff between passes straightforward because the working script, assigned voices, and project context remain together. You spend the more deliberate render on text that has already earned it.
Voices stay reusable across engines
One thing that trips up new users: the voice you pick isn't tied to a single workflow. The 100+ production voices in the voice library work across the core engines, and Pro users can route those voices through Omni for the broadest language coverage without re-selecting anything.
The same voice will sound slightly different across engines. That's expected. The fast engine produces a cleaner, more neutral read. The expressive engine adds more dynamic range and emotional depth. The multilingual engine adapts the voice to the phonetic patterns of its supported languages. Omni handles the broadest language coverage and voice design on Pro.
Think of the engines as different directors working with the same actor: the same voice can give a different performance in a different context.
Switching engines mid-project
You don't have to commit to one engine for an entire project. Switch freely. Plenty of creators use the fast engine for narration-heavy sections and the expressive engine for dialogue-heavy chapters. Some even generate English content with the expressive engine and localized versions with the multilingual engine, all within the same project.
The engine selector is in the toolbar. Two clicks, and you're generating on a different engine. Your script, voices, and project structure stay exactly the same.
Choose, compare, and keep the project moving
Start with the engine that fits the first problem in front of you, then review a representative section in context. Fast is the practical starting point for English drafts; expressive earns its place when delivery carries the story; multilingual and Omni cover localization needs.
Browse and preview candidates in the Vois voice library, then keep your approved script, voice, and engine choices together in the project. Get started when you are ready to test the workflow with your own material, or see current plan options.
The Vois Team