KittenTTS: Three New Open-Source TTS Models, Smallest Under 25MB

GitHub / KittenML·medium signal

KittenML released three new text-to-speech models on GitHub (432pts, 160 comments on HN — highest point total in today's preflight data) with the smallest model weighing under 25MB, enabling full on-device inference without cloud round-trips. The size profile targets mobile and edge deployments where existing TTS models (Kokoro, Parler) are too large for practical embedding. High HN engagement signals strong practitioner demand for lightweight, deployable voice models outside the API tier.