How Software Mansion launched Reverb-TTS on X

Clones a voice from one short clip and speaks seven languages on-device

What Software Mansion launched

Software MansionNov 19, 2026

Introducing Reverb-TTS, a state-of-the-art voice-cloning text-to-speech model for local inference. 🔊 Reverb-TTS can clone your voice from one short clip and make it speak Polish, German, French, Spanish, Italian, Portuguese, & English, all on your phone. It is SotA in both https://t.co/6vRbwyJopG

326373153K971

Transcript

  1. 0:00Software Mansion presents a zero-shot voice cloning model.
  2. 0:12Reverb!
  3. 0:13Who put reverb on that?
  4. 0:15Turn it off!
  5. 0:17Relax, that's just what the model is called.
  6. 0:20All it needs is ten seconds of your voice.
  7. 0:24And after that it will say anything you type.
  8. 0:27It's really, really small too.
  9. 0:29It doesn't even have 300 million parameters.
  10. 0:32And it's so fast that I can't keep up with it.
  11. 0:35The weights are open too, dear.
  12. 0:37So help yourself.
  13. 0:39Any voice! Any text!
  14. 0:42With Reverb TTS!
  15. 0:45Reverb TTS.
  16. 0:47Available now on Hugging Face.

How LaunchVideo.com rated the launch

LaunchVideo.com rating

2.80
Hook
Editing / Creativity
Copy
Sentiment of launch
Distribution strategy

How the launch performed

53K
views on X
#1204
of 1658 launches we track
0.3x
the 212K median launch

What this launch shows

Views for the Software Mansion launch compared with median views in the directory
Views compared with directory medians
GroupViews
Software Mansion, this launch53,371
Similar-sized accounts, median across 274 launches55,479
AI voice & audio launches, median across 66 launches384,803
Every launch, median across 1658 launches211,764

Software Mansion reached 53K views, 0.3 times the directory median of 212K, ranking 1204 of 1658 launches we track. Among AI voice & audio launches it ranks 46 of 66, 0.1 times that group's median. The post came from an account with 562 followers and reached 95 times that many people, ranking 143 of 274 launches from accounts of a similar size.

Replies ran at 6.0 per 10,000 views, well above the 3.7 median. Conversation in the first hours is what the ranking model rewards with more reach. Bookmarks came in at 181.9 per 10,000 views against a median of 14.8. One bookmark for every 0.8 likes is a high ratio and usually means people intend to come back and try the product.

The launch video runs 51 seconds, against a directory median of 74 seconds. It went out on a Thursday.

About the launch

Reverb-TTS is a zero-shot voice-cloning text-to-speech model from Software Mansion AI Labs, the Kraków company's group for models that run on phones, in the browser, and on other local hardware. A developer gives it 3 to 10 seconds of one speaker, plus a word-for-word transcript of that clip, and it reads new text in that voice in English, Polish, German, French, Spanish, Italian, or Portuguese. The same voice can switch languages, and the clip and the generated audio can stay on the device. This is the lab's first model, already packaged in react-native-executorch and in ruNNtime, the company's WebGPU engine for the browser. The model has 297 million parameters, plus a 22 million parameter codec, which is small enough to run on a phone. On the Open TTS Leaderboard's MiniMax set, the lab reports a 1.03 percent word error rate, lower than the larger open models in its published table, including OmniVoice, the open model its weights were initialized from. Speaker similarity in that comparison is 0.817, slightly under OmniVoice. The same write-up reports more than 20 times real-time generation on an Apple M4 Pro and an iPhone 17 at batch size 1. Weights are on Hugging Face under CC BY-NC 4.0, for non-commercial use with attribution. The inference code is Apache 2.0. Software Mansion, founded in Kraków in 2012, is known for React Native libraries such as Reanimated and Gesture Handler, and for tools that run models on device. Reverb-TTS applies that local-inference work to a speech model the lab trained, for app teams that want cloned voices without a cloud round trip. The project documentation asks users to clone only voices they have the right to use.

Tags

AI & MLAI voice & audioParody<500KB2D (developers)Product launchOpen sourceEU

Ask AI about Software Mansion

Questions

Frequently asked questions

The Software Mansion launch post reached 53K views on X, with 731 likes, 63 reposts, 32 comments, 971 bookmarks.

Software Mansion launched on X on November 19, 2026 with the message "Clones a voice from one short clip and speaks seven languages on-device".

Reverb-TTS is a zero-shot voice-cloning text-to-speech model from Software Mansion AI Labs, the Kraków company's group for models that run on phones, in the browser, and on other local hardware.

LaunchVideo.com rated the Software Mansion launch 2.80 out of 5: hook 1/5, editing / creativity 2/5, copy 5/5, sentiment of launch 4/5, distribution strategy 2/5.