A developer released SteerTTS, an English text-to-speech system built on a 9.6M-parameter acoustic model (attention plus bidirectional Mamba-3, FastPitch-style) paired with a Vocos vocoder, totaling roughly 23M parameters, that runs in real time on CPU. The developer said the model was built because no small, steerable TTS option existed for adjusting speech contextually from LLM-provided controls during a conversation, and SteerTTS accepts numeric z-score controls for pitch, rate and loudness.
I was looking for small TTS model which is steerable but couldn't find any, specially something I can adjust contextually by providing controls from LLM during a conversation. So I built one. What it is: 9.6M-param acoustic model (attention + bidirectional Mamba-3, FastPitch-style) + Vocos vocoder =
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.