Vois Studio 2.6 Guide: Six Engines, One Project

Last updated 20 October 2026 for Vois Studio 2.6.0.

Vois Studio 2.6 has six voice engines, and you pick one by the job. Sonic makes quick drafts on every plan. Fast handles revisions in five languages. Expressive adds English reaction tags like [laugh]. Multilingual speaks 23 languages with library voices. Craft directs any voice with a recording. Omni takes plain-word direction, designs new voices and can sing.

This is the written companion to our full walkthrough video. It follows one story: Juno, a solo creator, has one night to make a launch episode for her friend Otis's ramen bike, and Otis keeps adding requests. Juno and Otis are fictional. Every screen and every sound from the app is real.

What changed from Vois Studio 2.0 to 2.6?

Version	What you can do now
2.0	Pro plan with the Omni engine and Voice Design (describe a voice in words)
2.2.3	Encrypted backup and restore to one .voisbak file
2.3.1	Repeatable takes: the same text, voice and seed give the same audio
2.4.0	Six new speakers (Ellis, Mara, Caleb, Nora, Jules and Reed), 112 core voices, language and gender filters
2.5.0	New engines Sonic and Craft; Fast in five languages; Free that never expires; agent speech with vois speak
2.5.1	Each word highlighted as it's spoken; timeline edits kept when you regenerate; exports that hit the loudness you chose; steadier Craft voices
2.5.2	Real waveforms; switch a line back to an earlier take; line locking; backups restore across Mac and Windows
2.5.3	Whole-audiobook exports, including WAV files over 4 GB; Cancel stops an export within about a second
2.6.0	Voices you buy (from the Voice Store, coming soon) appear without waiting; a one-time email confirmation if bought voices are missing

One change to know about: in 2.5.0 Omni moved to 30 languages, fewer than before, but it's a much better engine for it. It reads more fluently, holds steady through long scripts, and you direct it freestyle in your own words; the old Omni only took a short, fixed set of bracket tags. Designed voices made only from a description need to be re-created; designs with a saved recording keep working. Every 2.5 change, release by release: What's new in Vois Studio 2.5.

Which Vois Studio engine should I use?

Engine	Voices	How you steer it	Languages	Use it when
Sonic	10 built-in voices	No tags or direction	31	Quick drafts, simple reads, agent speech. On every plan
Fast	Library, cloned and saved designed voices	No tags or direction	5: English, German, Spanish, Italian, Portuguese	The words are still changing and you want to revise quickly
Expressive	Library, cloned and saved designed voices	9 reaction tags	English	English dialogue needs an audible laugh, sigh or gasp
Multilingual	Library, cloned and saved designed voices	No tags or direction	23	Narration with one library speaker across many languages
Craft	Any voice with a recording: library, clones with a transcript, saved designs	Plain-word direction	10	Directing a voice you already have. Doesn't sing
Omni	Library, cloned and designed voices	Plain-word direction and Voice Design	30	Directed performances, new voices, singing

Sonic is on every plan. Fast, Expressive and Multilingual come with Subscriber and Pro. Omni, Craft and Voice Design are Pro; on Free you can try Omni and Craft and export their audio with purchased credits. Craft needs about 9 GB of memory on a Mac and 11 GB on Windows. More detail: Choosing the right voice engine.

Where do I switch engines?

1. Click the engine name in the top toolbar. Voice Settings opens.
2. Under TTS Engine, choose Sonic, Fast, Expressive, Multilingual, Craft or Omni. Each row shows its language count.
3. If an engine isn't downloaded yet, use the download button beside it. The panel also tells you if this computer needs more memory for it.

The choice applies to the audio you generate next, not to a single line. When a project uses several engines, generate one part, lock the lines you like, then switch.

How do I write a script for an AI voice?

Write for the ear. A voice reads exactly what you give it, so the script should sound like someone talking. The full primer: How to write a script for AI voices.

Keep each sentence to one idea, and mix short sentences with a few longer ones so it doesn't sound clipped. Read the script aloud once: if you run out of breath, so will the voice. Talk to someone. Contractions, plain words and concrete examples sound better than formal prose.

Punctuation shapes rhythm. A full stop ends a thought, a comma keeps it moving, an ellipsis gives a small hesitation. That's a rough nudge, though, never an exact timing. Quotation marks are just punctuation too; they don't switch the voice or act anything out. For exact silence, use pause pills. Markup tags like <break> aren't supported.

Speakers and structure

Type / in the editor to open six commands: Speaker, Pause, Heading, Title, Section and Note. Headings, titles, sections and notes are for you and are never spoken.

1. Type /, choose Speaker, type a name such as Host and press Enter. A coloured pill appears.
2. Every line after it belongs to that speaker until you add another. Repeat for a Guest.
3. Click a pill to change it.

Pasted plain text works too:

[Host]: It's eleven at night. The city is tired.
[Guest]: People ask if it's a food truck. It's a food bicycle.

Exact pauses

Type /, choose Pause and pick 0.3, 0.5, 1.0, 1.5, 2.0 or 3.0 seconds. Click the amber pill to fine-tune it with a slider from 0.3 to 3 seconds. In plain text, a pause is [pause=1000] (milliseconds).

Numbers, dates and abbreviations

Write them the way you want them said.

Instead of	Write
11pm	eleven at night
03/04/2026	April third, twenty twenty-six
$12.50	twelve dollars and fifty cents
API	A P I
Chapter IV	chapter four

More examples: How to write numbers, dates and acronyms for AI voices.

Names the voice gets wrong

Use the pronunciation dictionary:

1. Settings > Pronunciation.
2. Under Where new rules apply, choose This project only or Every project. A project rule wins over an every-project rule.
3. Add, then fill in Word (for example Koenji) and Pronunciation (Koh-en-jee), and Add.

Rules apply the next time you generate; existing audio isn't changed. Import CSV and Export CSV move a whole glossary. See the pronunciation dictionary guide.

Tags and directions

Both start with an open bracket, [, and the engine decides which you get:

• Reaction tags are for Expressive: [laugh], [chuckle], [gasp], [sigh], [groan], [cough], [clear throat], [sniff], [shush]. They're sounds, not moods. Fast, Multilingual and Sonic remove them before speaking, so the bracket text isn't read aloud.
• Directions are for Omni and Craft: plain words about tone, pace, emotion and energy, such as "quiet, close to the microphone, like sharing a secret". In plain text: [direction: calm and warm]. A direction lasts until the next direction in that speaker's turn. On a line spoken by another engine, the direction stays in your script but appears muted.

Guides: Expressive reaction tags and Direct an AI voice in plain words.

More than one language

Write each line in the language you want spoken; Vois Studio doesn't translate your script. The language picker on Auto detects the language. For a very short or mixed line, choose the language yourself. A cloned voice keeps the language and accent of the recording it was made from.

Sonic: how do I make a first draft?

1. In Voice Settings, choose Sonic.
2. In the Speakers pane, click Assign next to each speaker and pick a built-in voice: Cedar, Harbor, Slate, Ridge, Ember, Coral, Meadow, Juniper, Sage or Willow.
3. Click Generate Audio on the timeline and listen.

What Free includes: Free never expires. It includes Sonic, 10 generations a day, one project, three Listen items and 10 exports a month of English Sonic audio. You can try every other engine and listen; exporting their audio on Free uses export credits you buy. Settings > License shows how many Free exports are left this month. Subscriber adds Fast, Expressive, Multilingual and voice cloning; Pro adds Omni, Craft and Voice Design. See pricing.

Fast: how do I revise quickly with real voices?

1. Choose Fast in Voice Settings.
2. In Speakers, click Assign, use Search voices..., press the play icon to preview, then pick a voice (Juno used Nora and Caleb).
3. Fast speaks English, German, Spanish, Italian and Portuguese. Add a line in one of those, such as "Abrimos a las once. Trae hambre.", and click Generate Audio.
4. Edit a word and the clip shows Will be regenerated: text changed. Generate again; only that line is remade.

Expressive: how do I make an AI voice laugh?

1. Choose Expressive (English only).
2. Click where the sound should happen, usually just before the words it reacts to, and type [.
3. Choose from the list, for example Laugh.
4. Generate Audio.

[Guest]: [laugh] People ask if it's a food truck. It's a food bicycle.
[Guest]: [sigh] On a good night? Eighty.

One sound per spot works well. There's no [happy] or [sad]: put the feeling in the words. All nine tags and where to place them: Make an AI voice laugh.

Multilingual: how do I use one speaker in another language?

1. Open Voices > Library and use the language filter. Many library speakers have a recording in each of Multilingual's 23 languages; press play on a card to hear that language.
2. Choose Multilingual in Voice Settings.
3. Write the line in that language, for example 夜十一時、開店です。, and generate.

Making a dubbed track for each language of a YouTube video? See Add AI-voiced audio tracks to a YouTube video.

Craft: how do I direct a voice I already have?

1. Choose Craft (Pro; on Free you can try it).
2. On the line, type [. A direction chip opens.
3. Describe the delivery: "quiet, close to the microphone, like sharing a secret".
4. Generate Audio, then change the direction and generate again to hear the same voice perform it differently.

Craft works with any voice that has a recording: library voices, your clones that have a transcript, and saved designs. It has no built-in voices. It follows directions but doesn't sing. More examples and limits: Direct an AI voice in plain words.

Omni: how do I design a new voice?

1. Choose Omni in Voice Settings (Pro).
2. Voices > Add voice > Design Voice.
3. In Describe the voice, say who it is and how they sound, or click an example (Elderly Man, Young Woman, Deep Narrator, Soft-spoken, Elderly Woman, Child, Laid-back) and edit it. Juno wrote: "a woman in her thirties with a soft, warm, slightly husky voice, relaxed and gentle, like a folk singer at a small cafe."
4. Add Preview text, click Generate Preview and listen. Each preview is a little different; generate a couple.
5. Save Voice, name it, and it's in your voice picker.

Start from an existing voice keeps the same speaker and changes only how they speak: pace, emotion, tone.

Can Omni sing?

Yes, in our test. Here's exactly how we did it:

1. Add a speaker called Singer and assign the designed voice.
2. Write the lyric as the line: "Night shift noodles, steam on the window, one more bowl before the city sleeps."
3. At the start of the line, type [ and write: Sing this lyric as a slow, gentle folk song, with soft acoustic guitar playing behind the voice.
4. Generate Audio.

Omni sang the lyric and added backing music that matched the instruction. Every take is a little different. If a take speaks instead of singing, make the instruction plainer and generate again. Our fallback: Sing these words as a melody, not spoken. A slow, warm folk song with soft acoustic guitar. Short, rhythmic lyrics work better than paragraphs. Full walkthrough: How to make an AI voice sing from text.

How do I keep a good take?

• Earlier takes: right-click a clip > Takes lists earlier takes with their time and length (it appears once a line has more than one take). Choosing one switches back and locks the line; Undo on the message reverses it.
• Lock a line: right-click > Lock line, or click the clip and press L. Generate Audio keeps a locked line's take. A lock shows beside the line in the script; click it to unlock.
• Get another take: change the text or the direction and generate again.
• Word highlighting: as audio plays, each word is underlined as it's spoken, in the editor and in Listen.
• Waveforms: clips show their real waveform, so silences and cut-off endings stand out.

How do I edit the timeline?

1. Tracks: generated lines land in script order on one Dialogue track, each clip in its speaker's colour; music and effects get tracks of their own. Click a track and press M to mute it.
2. Move and trim: drag a clip to move it; drag its edges to trim.
3. Clip keys: click a clip, then P preview, S split at the playhead, T exact trim (Source offset, Clip duration, Apply trim), L lock, Delete remove. Space plays the timeline.
4. Clip menu: right-click for Duplicate, Reset Length, Edit Trim..., Fade In and Fade Out (None, 0.1s, 0.2s, 0.3s, 0.5s, 1s or Custom...).
5. Music: Add Media > Add Media Track > Built-in Library > Music > Ambient > Soft Ambient Background Music > Add (or Get, then Add). Lower the track's volume slider under speech.
6. Sound effects: same library, switch the type to SFX. Your own files: Import File (MP3, WAV, OGG, FLAC, M4A, AAC).
7. Undo: Cmd+Z on a Mac, Ctrl+Z on Windows. Timeline undo doesn't undo script edits.
8. Regenerate safely: change a line and press Generate Audio. Only changed lines are remade; the rest keep their trims, splits, fades, volume and spacing, removed lines stay removed, and gaps are kept even if you moved clips yourself.
9. Failures: the count next to Generate Audio lists every failed line with its reason; click one to jump to it. A moved music file shows File not found with its name.

Every timeline control in one place: The Vois Studio timeline, step by step.

Which export settings should I use?

Click Export on the timeline, then choose a Mastering Profile:

Profile	Loudness target	Peak limit	Format
YouTube Video	-14 LUFS	-1 dBTP	AAC, 48 kHz
Apple Podcasts	-16 LUFS	-1 dBTP	AAC
Spotify Podcasts	-16 LUFS	-2 dBTP	MP3, 192 kbps
ACX Audiobook	-20 LUFS	-3 dBTP maximum	MP3, 192 kbps, 44.1 kHz
High Quality	-19 LUFS	-1 dBTP	WAV, 24-bit, 48 kHz

LUFS measures how loud audio feels. With Loudness Normalization on, exports reach the target within 0.3 LU and stay under the peak limit; if a recording can't, the export says so. Prefer your own settings? Choose Custom and set Output Format (MP3, WAV, FLAC, AAC), Sample Rate (44.1, 48 or 96 kHz) and bitrate or bit depth; the dialog only offers combinations the format can save. If a line is missing audio, the export stops and names it. Every export carries an inaudible watermark. Background: Understanding LUFS, Export presets.

How do I back up my projects?

1. Settings > Backup > Create a backup.
2. Tick Include generated audio to keep your takes.
3. Create backup file..., choose a passphrase of at least 12 characters, confirm it, and Create backup.

You get one encrypted .voisbak file with projects, scripts, voices and audio. Licences and credentials aren't included. Restore from a backup brings it back on a Mac or Windows computer, whichever made it; it replaces what's on that computer and keeps a rollback copy first. Step by step: Back up and restore Vois Studio.

Bought voices missing? The Voice Store, coming soon, is a store inside Vois Studio where you can buy extra premium voices: a one-time purchase, and the voice is yours to keep. The voices your plan includes stay included. Voices you buy, whether in the Voice Store, on another computer or through a partner offer, need a one-time email confirmation. In 2.6.0, if they aren't in your Library, it shows "Confirm your email to use the voices you bought". Click Confirm email, enter the 6-digit code we send, and click Confirm.

What hidden features are worth knowing?

1. Templates: Home > Start from Template: Podcast, Audiobook, Video Script, Training, Advertising, Game Dialogue.
2. Example scripts: in an empty script, Insert example script offers ten samples.
3. Import scripts: Script actions (three dots) > Import Script... or Import All Scripts... for Markdown and text files.
4. Jump from script to timeline: double-click script text.
5. Ruler keys: focus the ruler; arrow keys move 1 second, Shift+arrow 5 seconds, Home and End jump to the ends.
6. Keyboard clip menu: focus a clip and press Shift+F10.
7. Listen: turn text, files (.txt, .md, .pdf, .epub, .docx) or a web page into audio; click a word to jump there; play at 0.5x to 2x.
8. Pronunciation CSV: import and export your glossary.
9. Model quality: Settings > Quality > Model quality: Efficient or High quality, when the computer has the memory.
10. Shortcuts: Cmd+S saves the project; Cmd+plus and Cmd+minus zoom the app (Ctrl on Windows).

Each one with the exact steps: 12 Vois Studio features you might have missed.

Can I use Vois Studio from the command line or with AI agents?

Yes. vois-cli comes with paid plans:

vois-cli tts generate --text "The secret is the broth." --engine craft \
  --language English --instruct "quiet, like sharing a secret" --output close.wav
vois-cli status --format json

--format json gives scripts and agents one machine-readable result. On every plan, Free included, agent speech reads an AI agent's output aloud with a Sonic voice; nothing is saved and it doesn't count toward Free's daily limit:

printf 'Build finished. Two tests need you.\n' | vois-cli speak --follow --play

More: How AI agents control your voice studio.

Which engine for which job? A recap

If you need...	Use
A quick draft, on any plan	Sonic
Fast revisions in English, German, Spanish, Italian or Portuguese	Fast
A laugh, sigh or gasp in English	Expressive
One library speaker in 23 languages	Multilingual
To direct a voice you already have	Craft
Direction, a brand-new voice, or a song	Omni

Download Vois Studio and start with Sonic on Free.

The Vois Team