Vois Studio 2.5: Sonic, Craft and a Free Plan That Stays

Last updated 20 October 2026. This post covers Vois Studio 2.5.0 to 2.5.3.

Vois Studio 2.5 added two voice engines, Sonic on every plan and Craft on Pro, made Free permanent, and moved Omni to plain-word direction in 30 languages. Updates 2.5.1 to 2.5.3, released through 19 October, added word highlighting, kept timeline edits, real waveforms, earlier takes, line locking and long exports that finish.

Most voice projects start the same way. You want to hear the script out loud before you commit to anything: the voice, the pacing, whether the joke lands. Then you want to shape one line without losing the ones that already work.

Version 2.5 is built around both halves of that. There's a quick draft engine on every plan and a way to direct a voice you already have, and four releases made editing and exporting more dependable.

Added 20 October 2026: you can see all of it at work in our complete Vois Studio walkthrough. The version story starts here:

What's new in 2.5.0?

Released 11 October.

Sonic, a fast engine on every plan. Sonic is on every plan, Free included, and speaks 31 languages with ten built-in voices: Cedar, Harbor, Slate, Ridge, Ember, Coral, Meadow, Juniper, Sage and Willow. It's the engine for a first draft or a simple read. It doesn't use library, cloned or designed voices yet. It needs at least 2 GB of free memory on a Mac and 4 GB on Windows, and Vois Studio tells you when there isn't enough.

Craft, direction for a voice you already have. Craft is a Pro engine. It speaks any voice with a recording (library voices, your clones with a transcript, and saved designs) in 10 languages: Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian and Spanish. Describe how a line should sound, such as "quiet, close to the microphone, like sharing a secret", and Craft follows it. Cloned voices keep the accent of their recording. Craft needs about 9 GB of memory on a Mac and 11 GB on Windows. On Free you can try it, generate and listen; exporting its audio uses purchased credits. More in Craft and Omni style direction.

Free never expires. Free now includes Sonic and its voices, 10 generations a day and 10 Free exports a month, which reset on the 1st (UTC). The Free exports cover English audio made with Sonic. Purchased export credits cover other languages, other engines and exports beyond the monthly 10. Settings > License shows how many Free exports are left. Free is for non-commercial use (pricing).

Fast now speaks five languages: English, German, Spanish, Italian and Portuguese. If it can't tell the language of a line with confidence, it asks you to choose instead of reading the text as English.

Voice Design starts with a description. Describe the voice in your own words, or pick an example and edit it. Or start from an existing voice and change how it speaks (pace, emotion, energy) while it stays the same speaker. Each take sounds a little different, so generate again to compare before you save. The guide: Omni Voice Design.

Agent speech. Pipe an AI agent's output into vois-cli speak --follow --play and Vois Studio reads it aloud on your computer with a Sonic voice. It works on every plan once Vois Studio is registered, saves nothing, and doesn't count toward Free's daily limit. The CLI can also run Vois Studio without a window and manage your licence from the terminal. Details: the Vois Studio CLI and AI agents.

Smaller fixes I'm glad to see go:

• English speech reads clock times and phone numbers naturally: "3:30 PM" as "three thirty PM", and "555-0199" digit by digit.
• Exporting a long Listen item saves the whole recording, not just its first seconds.
• Settings > License names the plan you bought, such as "Lifetime Tier 2" or "Founder Pro".
• Pro customers whose verified email also holds an active AppSumo Tier 2 or Tier 3 licence get the AppSumo-exclusive voices on Pro too.

AppSumo lifetime plans keep their included engines and don't include Omni or Craft.

What changed for Omni?

This is the change I most want you to read before you update.

Omni now speaks 30 languages: Arabic, Burmese, Chinese, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Indonesian, Italian, Japanese, Khmer, Korean, Lao, Malay, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Tagalog, Thai, Turkish and Vietnamese. That's fewer languages than before. When you ask Omni for a language it doesn't speak, it now says so instead of trying. For those languages, Sonic (31 languages) or Multilingual (23) may cover you; the engine guide lists each one.

So why make the trade? Because the new Omni is a much better engine. It reads more fluently, and it holds steady all the way through a long script.

It also listens to you. The old Omni only understood a short, fixed set of bracket tags such as [laughter]. Now you direct it freestyle: describe the delivery in your own words, like "tired but trying to sound cheerful", and Omni performs it. Tags already in your scripts are left as you wrote them, so replace them with a direction where you want the effect.

Designed voices made only from a description need re-creating. The upgraded Omni can't reproduce designs that were only a description in the previous Omni. Vois Studio marks them and asks you to re-create them from their description before you use them again. Designed voices with a saved recording keep working.

What changed in 2.5.1?

Released 15 October.

• Word highlighting. As audio plays, the script highlights each word as it's spoken, in the editor and in Listen.
• Regenerating keeps your timeline edits. Generate Audio keeps trims, splits, fades and volume on unchanged lines, keeps removed lines removed, and keeps the gaps between clips, even when you've moved clips yourself. Changed lines play their full new take. More in multi-track timeline editing.
• Exports hit the loudness you chose, within 0.3 LU, and stay under the peak limit. Before, they came out 2.5 to 3.3 LU quieter than asked. If a recording can't reach the target without going over the limit, the export says so.
• Export settings are honoured. WAV and FLAC save at the bit depth you pick, MP3 and AAC at the bitrate you pick, and each platform profile saves in that platform's own format. Long high-resolution exports are much faster: an 11-minute episode at 96 kHz now exports in about 20 seconds instead of nearly 2 minutes. See the export presets guide.
• Craft voices stay steady through a line and sound closer to the voice you chose. Takes you already made keep their audio; generate a line again for the improved version.
• Lines that stall or get cut short are regenerated on the same engine or reported as failed, instead of keeping incomplete audio or a long silence.
• Missing audio is marked on the timeline and regenerated with Generate Audio. Exports stop and name the line instead of leaving a silent gap, and moved or deleted music and effects files are listed by name.
• The failed count beside Generate Audio opens a list of failed lines, each with its reason, and takes you to a line when you choose it.
• Text ending in a quotation or containing an ellipsis is read as written, without stray sounds or extra pauses.
• Vois Studio is a smaller download and takes less disk space. Quitting on macOS after generating no longer crashes, and tutorial videos open on YouTube in your browser.

What changed in 2.5.2?

Released 18 October.

• Earlier takes. Right-click a clip, then Takes, to switch a line back to an earlier take.
• Line locking. Lock a line so Generate Audio keeps its audio: click the clip and press L. Choosing an earlier take locks the line for you, and Undo puts it back.
• Real waveforms on every clip, so silences and cut-off endings are visible at a glance.
• Backups move between Mac and Windows. A backup made on a Mac restores on Windows, and the other way round. See back up and restore Vois Studio.
• Activating on a new computer shows which inbox to check when Vois Studio needs you to confirm the device by email, lets you resend it, and finishes once you select Confirm this device.
• Short English scripts of a sentence or two are recognised as English with Auto language.
• Clip shortcuts act on the clip. After you click a clip, P, S, T, L and Delete act on it instead of typing into the script.
• Creating a project past Free's one-project limit explains the limit and offers Upgrade, instead of doing nothing.

What changed in 2.5.3?

Released 19 October.

• Whole-audiobook exports finish. Exporting a very long project saves the complete recording instead of stopping partway without an error. A WAV export larger than 4 GB is saved as RF64, the standard format for WAV files that size.
• Cancel works throughout an export. While the final file is encoded, watermarked or checked, Cancel stops the export within about a second and nothing is saved; it also takes effect while the audio is still being decoded. Long built-in library tracks such as "Ambient Background" now export too.
• One clear message when a voice can't be used. Generate Audio tries that speaker's lines once, says which speaker and what to do, and keeps generating everyone else's lines.
• Long scripts no longer slow down as lines finish, and the completed and failed counts are accurate: every failed line is listed, and only lines that got new audio count as completed.
• Damaged engine files are repaired. Vois Studio says when an engine's downloaded files are damaged and offers Download, which repairs it.
• When an audio file is there but can't be played, the timeline says "Can't read file" instead of "File not found".

Do I need to do anything?

Mostly no. A few things are worth a minute:

1. Re-create description-only designed voices. Vois Studio marks them. Open each one and re-create it from its description.
2. Check Omni projects in other languages. If a project used a language Omni no longer speaks, choose an engine that speaks it.
3. Swap Omni bracket tags for directions. Replace a tag like [laughter] with a plain-word direction on the line.
4. Regenerate Craft lines you want to sound steadier. Existing takes keep their audio.
5. Expect fresh takes in two places. English text with a time or phone number, and text you generate again on Fast, makes a new take the first time.

Try it in five minutes

1. Draft with Sonic. Choose Sonic in Voice Settings, paste a paragraph, and click Generate Audio. Watch each word highlight as it plays.
2. Direct one line. On Pro, or trying it on Free, switch to Craft, type [ on a line and describe the delivery in plain words. Generate and compare.
3. Keep the take you like. Right-click the clip, open Takes, choose the better one, and it's locked for you.

Download Vois Studio to get started, or read the full changelog for every 2.5 fix.

Related

• The complete Vois Studio 2.6 guide
• Which Vois Studio voice engine should I use?
• Craft and Omni style direction
• Multi-track timeline editing
• Export presets guide
• Back up and restore Vois Studio
• The Vois Studio CLI and AI agents
• Vois Studio 2.4: More natural speech, a bigger cast

I'm proud of this one for a simple reason: you can start on Free, keep going as long as you like, and the work you shape on the timeline now stays shaped.

-- Praney