Six new speakers. 112 core voices.
Ellis, Mara, Caleb, Nora, Jules and Reed join the cast, bringing the core library to 112 voices. More choice for your next narrator, podcast guest or character.
Release notes
Version 2.4.0 ·
More natural multilingual speech
Your built-in voices already spoke multiple languages, but an English accent could carry into other languages. Vois 2.4.0 improves accent, rhythm and style across the 23 supported Multilingual languages, helping your chosen speaker sound more natural in your script’s language. You also get six new speakers, easier voice selection and fixes that keep your projects moving.
Improved delivery across 23 supported languages
Better delivery for your AppSumo voices too
The improvements to accent and style also apply to the 100 AppSumo voices if they are included in your licence. Your existing access stays the same.
Hear the language before you choose
Audition your chosen speaker in the language of your project before generating a full passage. Language filters in the Voice Library, Favorites and website voice browser make it easier to compare voices.
These improvements apply to built-in voices and the AppSumo collection included with your licence. Cloned voices retain the language and accent characteristics of the recording used to create them. For the most natural delivery, use a voice cloned in the language of your script.
Clearer pronunciation, more complete narration
Corrections across languages and writing systems
Fewer wrong words in affected Mandarin sentences, more accurate Arabic word forms and improved pronunciation in Greek, Turkish and Korean. Several Japanese readings are corrected too.
Longer passages stay on track
Longer passages are less likely to stop early, repeat phrases or stray from your script. Narration in German, French, Hindi, Italian and Japanese follows the complete text more reliably, with fewer long, unwanted pauses.
Non-English text stays in its own language
Numbers, prices and abbreviations are spoken more consistently in the selected language instead of unexpectedly switching to English. Arabic narration with Omni is also more reliable.
Preserve the sound through playback and export
Louder Omni passages are less prone to distortion. Fixes to WAV export and copy preserve the quality of your generated audio.
These corrections are not a promise of native-speaker pronunciation. Individual words can still be read unexpectedly, including Chinese words with multiple readings, Hebrew without vowel points and some Japanese text. Listen to generated speech before publishing.
Choose the language once, then keep working
Listen detects your text’s language
Listen can choose the language automatically. Use the manual override when you need a particular language, especially for short or ambiguous text.
Language controls where you need them
Choose the language beside Save in the editor. When working through the CLI, let Vois detect it or set it yourself for short passages, mixed-language scripts or romanized text.
Keep CLI speech running while you edit
Switch projects or leave the editor without interrupting speech started through the CLI. Previews also work more reliably for saved cloned voices.
A refreshed interface, with voices easier to find
Find male and female voices more easily
Filter by male or female voice alongside language, category and search in the Library, Favorites and website voice browser. Favorites now follow your filters correctly, and included AppSumo voices are easier to identify.
A recognizable face for every built-in voice
Each built-in voice now has its own personalized avatar, with male and female voices represented correctly. The same avatar appears in the app and on the website, so your chosen speaker is easier to recognize wherever you browse.
Choose an avatar for the voices you create
Choose an avatar when cloning a voice or saving a voice made with Voice Design. Your choice follows that voice throughout the app. You can also set and edit its language and gender labels. Cloning and Voice Design remain subject to your existing plan access.
A tidier workspace for everyday decisions
A more compact Library toolbar brings filters and sorting together in card and list views. Audiobooks and Podcasts are easier to find, and the editor’s language choice sits beside Save.
Clean up several voices together
Select and delete multiple cloned or designed voices together, including a filtered selection. Review your choice before deleting, and retry any voices that could not be removed.
See which quality setting is selected
Fixed the legibility issue that made the selected model quality hard to see. Efficient and High quality now have a clearly visible selection in both light and dark themes, with keyboard support.
Less guesswork when something needs attention
Clearer help when a voice is unavailable
Clearer messages explain when a voice download has failed or the selected language is unsupported, so you know whether to retry or choose another language.
More reliable generation and voice cloning
Fixed generation failures after computer cleanup and improved voice-cloning reliability when optional notes are left blank.
Keep licence updates from interrupting access
Licence refreshes no longer incorrectly sign you out or interrupt access to your voices.
Find your next workflow
A clearer introduction to Vois
Explore audiobook, podcast and voiceover workflows on the rebuilt homepage, watch the demo, and find voices more easily with language and gender filters.
Open the tutorial you actually selected
Tutorial cards now open the video you selected. Updated guides help you choose languages, direct speech and create multilingual projects through the app or CLI.
Earlier improvements: 2.3.4 and before
These desktop improvements were already included in the previous release notes. They remain part of Vois, but are not new in 2.4.0.
A verified publisher on Windows
The Windows installer and app identify SEED BLOCKS PTY LTD as the verified publisher, so you can check who supplied your download.
Better Japanese pronunciation
More accurate readings for Japanese text containing kanji.
Listen imports return to the right place
Web imports no longer appear to fail when they finish quickly. After importing, you return to the editor correctly, and saved Listen items keep their original source details.
Project switching shows the current selection
Switching between projects or scripts no longer shows content from an earlier selection.
Music and sound effects keep their placement
Music and sound effects fit the narration when first added, then keep their placement and duration when you edit or regenerate narration.
More reliable waveforms and playback
Waveforms remain usable on long timelines. Playback no longer jumps to an earlier position or restarts after you stop it when controls are used in quick succession.
Clearer stopping points for large generation jobs
If your computer does not have enough available memory to begin, Vois stops the job with an explanation instead of repeating the same error for each remaining clip.
Clearer guidance when an export needs retrying
If an audio file changes while you are exporting, Vois explains that you need to retry.
Safer retries after an interrupted export
Retrying an interrupted export no longer risks deducting the same credits twice.
Fewer interruptions to long CLI jobs
Longer multilingual generation jobs can finish without the CLI ending its wait too early. This improves reliability, not generation speed.
Earlier website and onboarding improvements
These download and onboarding changes also predate 2.4.0.
Correct downloads for Apple Silicon Macs
Apple Silicon Mac users are more reliably offered the correct download, including when using privacy-focused browser settings.
Older activation links remain useful
Existing activation links continue to provide the licence key and current download instead of leading to a dead end.
Update to Vois 2.4.0
Update Vois on your Apple Silicon Mac or Windows PC. Stay online while the app prepares your voices for first use. Preview your chosen speaker in your script’s language, and listen to the finished audio before publishing.