Vois 2.0 was built for the point where a good voice is not enough. Creators also need to revise a line, keep a recurring narrator consistent, finish the mix, and export without sending the script through a cloud workflow.
The update puts that work in one local studio: choose a voice, build the script, generate, arrange, master, and export. If you need a custom narrator or a multilingual version, those choices now sit inside the same project rather than becoming a second tool chain.
Vois remains desktop-native and local. The new release adds more ways to make the outcome fit the project, while keeping scripts, audio, and voice samples on the machine where you are working.
Choose the engine for the job
Vois 2.0 gives you distinct voice engines rather than one compromise setting.
Fast is for English drafts and quick revisions. It generates at about three times real time on a standard CPU, and on supported Apple Silicon Macs it can reach about six times real time with GPU acceleration.
Expressive is for English work where delivery carries the meaning: narration, characters, podcasts, and finished voiceover.
Multilingual covers 23 languages for projects that need a consistent workflow beyond English. Pro adds Omni, which supports 600+ languages and Voice Design for making a new voice from selected attributes.
That choice is useful because a first pass and a final pass do not have the same needs. Draft quickly, then choose the voice and engine that give the final recording the right pace and character.
100+ voices across 21 categories
The library now has 100+ voices organized around the roles people actually cast: narrators, hosts, broadcasters, educators, storytellers, companions, characters, and game voices. Start with the category that suits the script, preview a few candidates, then keep the winning choice attached to the project.
The practical outcome is continuity. A YouTube channel can keep one recognizable narrator. A training team can keep an instructor voice across updated modules. A fiction project can give each recurring character a stable identity without rebuilding the setup for every chapter.
Voice cloning with a local review loop
Voice cloning turns a short, clean recording into a reusable custom voice. Aim for about 15 seconds of natural speech, make sure you have the speaker's permission, and create the clone in the app. The sample is processed locally and the clone stays with your local library.
Do not treat the first result as final. Preview it with a real sentence from the project, listen for the right pace and clarity, and replace the reference sample if the delivery misses the mark. Once the clone is ready, assign it to a speaker, generate the script, and review the result on the timeline alongside the other voices.
Faster local generation on Apple Silicon
The Fast engine is designed for the revision-heavy part of production. On Apple Silicon Macs with GPU acceleration, it can reach about six times real time. That gives you room to test a different sentence, compare two voices, or rebuild a section without waiting on an upload and download cycle.
Other supported machines can still use the Fast engine on CPU. The point is not a benchmark for its own sake. It is a shorter gap between noticing a problem and hearing a better version.
Let an agent handle repeatable production
Vois 2.0 includes a local CLI for repeatable production work. An AI agent can use the hosted Vois skills to create projects, prepare scripts, assign voices, generate batches, and export using a mastering profile. The app remains the local studio that performs the work.
Prompt your agent: "Read the Vois CLI skill at https://vois.so/skills/vois-cli/SKILL.md. Create a local project for this script, propose voice assignments and an export profile, but show me the plan before generating or exporting anything."
Expected deliverable: a project plan, proposed voice IDs, a script structure, and the intended export destination.
Review and approve:
- Confirm the project name, speaker assignments, and script text.
- Preview one short sample per speaker in Vois.
- Approve the batch generation only after the previews sound right.
- Listen to the mastered export before sharing or publishing it.
What changed in the writing workflow
Vois 2.0 centers the writing workflow on choices the producer can hear and review. Choose a voice that fits the part, use a clone only for an authorized specific voice, and shape delivery with ordinary punctuation, paragraph breaks, and the timeline.
The edit loop stays clear: revise the words, audition the result, then keep or regenerate the affected section.
Plans and existing work
Subscriber covers the core local studio, including the voice library, voice cloning, local engines, CLI access, and export. Pro adds Omni and Voice Design. Plans and entitlements change over time, so see the current details on pricing rather than relying on an old announcement.
For existing projects, review scripts, speakers, timeline arrangements, and custom voices after migration before using them in new work. A quick sample export is the safest way to confirm that the project still sounds as intended.
Start with one finished minute
Open a small real project. Pick a voice, make a 30 to 60 second sample, arrange it on the timeline, apply the destination's export preset, and listen back. That single minute tells you more about the new workflow than a feature list can.
For a series, do this with the hardest minute rather than the easiest one. Use a segment with names, transitions, two speakers, or an unfamiliar word. It tests the real production path: voice choice, pronunciation, timeline handling, mastering, and export. When that minute works, you have a reliable pattern for the rest of the project instead of a feature checklist that only sounds impressive on paper. It also exposes any weak links before the project becomes difficult to unwind: an unsuitable voice, an unclear pronunciation, a poorly chosen preset, or an approval step that needs a clearer owner.
Explore the Vois voice library or CLI and AI Automation, then Get started.
The Vois Team