What's new · Stentor

0.1.3 7 November 2026

Qwen3-TTS, a new model for English scripts: the fastest and most natural, about 3× faster than Turbo. It whispers; it has no moods or sounds.
Hear a voice before you pick it: every voice in the voice menu has a play button.
Language symbols: start a sentence with [spanish], [hindi] and the rest to read it, and the rest of its paragraph, in that language.
Help improve Stentor: send anonymous reports on errors and voicing speed, if you switch it on (Settings › Privacy). Never your scripts or your voice.
The script being voiced shows its progress in the sidebar.

Shorter pauses: 0.3 s between sentences and 0.5 s between paragraphs, the way a lively reader pauses.
Model is one list you can search, grouped by Local AI and ElevenLabs.
Short sentences in Spanish, French, German, Italian and Portuguese are found in an English script.

Picking an ElevenLabs model no longer leaves the model list loading.
No more "can't be voiced yet" for other languages when an English model is picked.

Restart to update installs the new version and opens it: Stentor could get stuck while quitting, so the update never installed.
Closing Stentor quits it fully.

41 built-in voices in 11 languages, made for Stentor: no recordings of real people.
A Voices page with one card for every voice, with filters and sorting.
VoxCPM2 for English scripts: the closest to your own voice.
Translate and Enhance with a writing model on your computer: no account, nothing leaves it.
A new welcome: the models download in the background while you record your voice, and the first voiceover starts by itself.
Skip the download on the welcome: write your scripts straight away, and download when you first voice one.
Search in the language and voice menus.
A Logs page in Settings › About, for looking into a problem.

The first public beta. Everything in it:

Write with moods, pauses, pace and sounds; type @ to pick one.
Voice it on your computer, with every word checked by Whisper and wrong words voiced again automatically.
Words light up as they play; click one to play from there.
Edit and voice again: only the changed parts are redone. Versions and earlier voiceovers are kept.
Selection toolbar: Play from here, Fix how it's said, Say it like this, Translate, Add symbol.
Enhance: your writing assistant adds symbols in the style you pick, with a before and after to review.
Export the voiceover, SRT captions, word timings and the script.
Voice one language at a time: just the changed parts, or the entire script.

Clone your voice with live, verified consent, and record your own whisper and sounds.
The script's language is found for you; mixed-language scripts are voiced sentence by sentence.
Hindi, Tamil and 9 more Indian languages with Maya2; Spanish, Arabic, French, German and 17 more with Chatterbox Multilingual.
One script, many languages: each translation is a tab of the same script, with its own voice and voiceover.
Change the English and the other languages show what's behind; Update translation redoes only those paragraphs.
Export every language at once, a voiceover and captions each.
Translate one selected sentence in place.
Recording starts on your computer's own microphone, and warns when a Bluetooth headset would sound muffled.

Choose the models you need in setup, with live download progress; add or remove them any time.
Claude Code and Codex as writing assistants, with your own plan.
ElevenLabs voices and models with your own key; the model list follows the script's language.
In setup, voice on your computer or with ElevenLabs: the local models are never forced.

Loads only the models the work needs and gives the memory back when it's done, so your computer stays fast.
Settings shows the memory in use and the models on disk, when each was last used, and clears working files.
Tells you when a new version is out, and installs it from the app.