March 19, 2026
Small cats, loud takes
Show HN: Three new Kitten TTS models – smallest less than 25MB
Tiny TTS drops — iPhone pleas, 'business-ready' beef, and belly-laugh tags
TLDR: Kitten TTS launched three tiny voice models that run on normal CPUs. The crowd loves the size but clashes over iPhone support, business-ready voices, emotion controls, and language options, with a few warning about the smallest compressed model—tiny tech, big expectations.
The devs behind Kitten TTS just released three teeny text-to-speech models that run on regular computers — no fancy graphics card needed — and the internet immediately turned into a talk show. Fans cheered the size-to-sound magic, calling the 25MB option “amazingly good,” while curious cats asked if it’ll work right in the browser and how to try it fast (hint: there’s a demo and a direct download).
Then the drama pounced. The top ask: “Can it run on iPhone?” With a mobile kit on the roadmap, the crowd is impatient and loud. The sharpest take came from folks trying to use it at work: one commenter slammed the built-in voices as “unusable in a business context,” pushing for DIY custom voices and clear pricing. Another camp wants acting lessons baked in, begging for more control and playful stage directions like “[laughs in melodic ascending and descending arpeggiated gibberish babbles]” for real emotion, not robot radio. Language support also got side-eyed: “Is it English only?” sparked calls for multilingual ASAP.
Meanwhile, a reviewer name-dropped a rival tool and praised Kitten’s quality-for-size while throwing shade at the voice lineup: “don’t love the voices, but it’s not bad.” Bonus spice: a few users flagged issues with the tiniest compressed model, adding a little “handle with care” energy to the launch. In short: tiny cats, big claws, louder opinions.
Key Points
- •Kitten TTS v0.8 releases 15M, 40M, and 80M parameter ONNX-based TTS models optimized for CPU-only inference.
- •Models range from 25 MB (int8) to 80 MB on disk, output 24 kHz audio, and include eight built-in voices with adjustable speech speed.
- •The library is in developer preview; APIs may change, and commercial support is available for integrations and custom voices.
- •Models are downloadable from Hugging Face Hub under KittenML, with an online demo on Hugging Face Spaces.
- •An int8-quantized 15M model (~25 MB) is available but may have issues; users are asked to open an issue if encountered.