Indie dev shrinks open-source TTS model Kokoro from 320MB to 40MB with voice intact
kingslyj · x · 2026-11-22
Indie developer nutanc used research agents to run overnight experiments compressing the 82M-parameter open-source TTS model Kokoro from its default 320MB down to 57MB, and eventually to 40MB while keeping most voice characteristics intact — even fitting a Hindi voice into the same 40MB file.
The earlier 57MB version showed slightly off pauses in listening tests, which the author says should be fixable; an 11MB compressed model is also in training. Relevant for anyone interested in on-device open-source speech synthesis.
More from coding & agent
- Dev circles revisit delegation: the era's task is trusting AI with more work — 4310sy · 2026-11-22
- Running Claude Design via Claude Code with third-party APIs using the Open Artifact project — dotey · 2026-11-22
- Garmin Health MCP Server lets you query VO2 max, sleep and training data via natural language — modelcontextprotocol · 2026-11-22
- Stripe MCP connector offers read-only access to Stripe data via API key — modelcontextprotocol · 2026-11-22
- Dev ports Inkling 2 Mini to vLLM, hitting ~200 tok/s decode on 2x RTX 3090 vs 50 on llama.cpp — peaster_ · 2026-11-22
- Six OAuth pitfalls from wiring Claude and ChatGPT into a remote MCP server — Dheeth · 2026-11-22