A voice can speak your script's language and still sound out of place. The words may be French, while the accent and rhythm lean toward English.
Vois already supported using the same speaker across languages. Version 2.4 improves how those built-in voices sound in the language you choose, with more natural accent, rhythm and style. Six new voices join the library too, alongside fewer interrupted or repeated passages and easier voice selection.
Vois 2.4.0 was released on September 21. This September 22 story follows my earlier preview. Then, I described the direction. Now I can show you what shipped.
More natural delivery in languages you already use
Imagine preparing a lesson in English and a second version in French. You've already chosen the narrator. You want the French version to sound natural in French, not like an English performance carried into another language. That's the improvement this release makes across the 23 supported Multilingual languages.
You can also audition each built-in speaker in your project's language before generating the full passage. Hear the accent, rhythm and character of the voice, then decide whether it suits your script.
This is a casting improvement as much as a language improvement. You can hear a candidate in the language of the project before committing to a longer read. You aren't choosing a voice from one performance and hoping it fits an entirely different brief.
You supply the script in the language you want spoken. Changing the language doesn't translate the words. Multilingual's 23-language workflow doesn't require Pro; existing access rules still apply. Omni's existing 600+ language support remains a separate Pro feature.
A note on cloned voices: these improvements apply to built-in voices and the included AppSumo collection. 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.
Meet six additions to the cast
Ellis, Mara, Caleb, Nora, Jules and Reed bring the core catalog to 112 speakers. All six arrive with the same 23-language coverage as the rest of the core library.
I like a new voice best when it changes a real casting decision. Put the same paragraph in front of two candidates and compare the read. Does this speaker fit the explanation, the story, the next episode? The new voices give you more choices without asking you to rebuild an established cast.
For owners entitled to the AppSumo collection, all 100 AppSumo voices get the same improvements to multilingual accent and delivery, plus language-specific previews. That collection remains tied to your existing licence, not unlocked for every account. Its badge and filter make it easier to find in the library.
Clearer pronunciation and more complete passages
Language coverage only matters if the generated passage follows the text. This release tackles accuracy, completeness and pacing together.
Affected Mandarin sentences have fewer wrong words. Arabic word forms are more accurate, pronunciation improves in Greek, Turkish and Korean, and several Japanese readings have been corrected.
The longer passage matters too. Multilingual narration is now much less likely to stop early, repeat a phrase or wander into unrelated speech. Multi-clause passages in German, French, Hindi, Italian and Japanese follow the complete text more reliably. Pacing fixes reduce excessive silence, so a sentence is less likely to turn into a long wait for its ending.
Numbers, prices and abbreviations are also spoken more consistently in the selected language instead of unexpectedly switching to English. These are the details that make a generated read more useful in an actual lesson, chapter or voiceover.
One editorial reminder: review the finished audio, especially names and language-specific readings. These fixes aren't a claim of perfect pronunciation or native-speaker certification.
A refreshed interface and a face for every voice
Find male and female voices with dedicated filters in the Library, Favorites and website voice browser. Language, category, search and sorting sit together in a more compact toolbar, in both card and list views. Favorites now follow your filters correctly.
Each built-in voice has its own personalized avatar, with male and female voices represented correctly rather than paired with a mismatched face. The same speaker has the same avatar in the app and on the website. Once you've found your narrator, it's easier to recognize them wherever you browse.
You can choose an avatar for a voice you clone, or when saving a voice created with Voice Design. That choice stays with the voice throughout the app. These features still follow your existing plan access.
You can label those voices by language and gender, then change the labels later. The labels describe the recording; they don't change its sound. And when experiments pile up, you can select and delete several cloned or designed voices together, including all visible filtered results, with confirmation and a retry option for partial failures.
The refresh also fixes a frustrating visibility problem in Settings: you can now clearly see whether Efficient or High quality is selected. Both light and dark themes make the selection readable, and the control works with a keyboard.
Let Listen and the CLI follow the work
Listen can detect the language of your text automatically, with a manual override when you want to make the choice yourself. In the editor, the language selector now sits beside Save and the script actions.
The CLI gains language selection for script generation and previews. Let Vois detect the language, or choose it yourself for short, mixed-language or romanized text.
There's also a particularly practical fix: changing episodes or leaving the editor no longer cancels unrelated speech started from the CLI. You can move around the studio without accidentally interrupting that separate job.
Try it with one paragraph
- Choose a familiar script. Use a paragraph already written in one of the 23 supported languages. Include the names or longer sentences your project actually needs.
- Audition in that language. Pick Multilingual, set the language and compare your current speaker with Ellis, Mara, Caleb, Nora, Jules or Reed using the matching-language previews.
- Generate the paragraph. Listen for the words, the ending and the pacing. Keep the speaker that serves the piece, then carry that choice into the rest of the project.
Download Vois to get started, or read the complete 2.4.0 changelog for the full release record.
I'm proud of this one because the headline and the everyday work belong together. A bigger cast is useful. Being able to find the right speaker, hear them in your script's language and finish the passage is the point.
-- Praney