Kitten TTS V0.8 is out: New SOTA Super-tiny TTS Model (Less than 25 MB)
Three new open-source models (80M, 40M, 14M parameters) run on CPU and match cloud TTS quality locally.
Deep Dive
Kitten ML has released Kitten TTS V0.8, an open-source text-to-speech model family under Apache 2.0. The three new models—80M, 40M, and a super-tiny 14M parameter 'Nano' version under 25 MB—offer eight expressive voices and major quality upgrades from previous versions. They are designed to run on CPU or resource-constrained edge devices, enabling free, local inference for on-device voice agents and applications without API calls.
Why It Matters
This enables high-quality, private voice features in mobile apps and IoT devices without expensive cloud APIs or GPUs.