Hi PH! We started with a pronunciation scoring API a few weeks ago and got great feedback. Since then we've expanded into a full speech AI platform. The insight: language learning apps and AI tutoring agents need pronunciation scoring, speech-to-text, AND text-to-speech together. Having all three from one provider, with consistent latency and a single API key, is surprisingly rare. The entire speech understanding stack (STT + pronunciation) shares a single 17MB model. TTS uses Kokoro-82M (115MB), which is currently #1 on the TTS Arena and Apache 2.0 licensed. We also ship an MCP server with 8 tools so AI agents (Claude, GPT, etc.) can use all capabilities as tool calls — no custom integration needed. Try the demo (3 tabs): https://huggingface.co/spaces/fa... Would love your thoughts on what features to add next!

Speech AI Platform

Pronunciation scoring, STT & TTS in 150MB — for AI gents

Pronunciation scoring, STT & TTS in 150MB — for AI gents