Open-source Verbalizer turns written text into speech-ready text in milliseconds on one GPU

EAVDR · reddit · 2026-11-20

Tontaube released TontaubeV1-Verbalizer under Apache 2.0, an LLM-based verbalization model that converts written text ($5, Dr., 9:05 a.m.) into clean spoken text for small TTS models. Fine-tuned from Qwen3 1.7B with custom tokenization, it handles 20,000 characters in under 2 seconds on an RTX 3090 (4-84 ms per sentence), reads contextually (Dr. = Doctor or Drive; IV stays IV for drips but reads Four for chapters), and copes with links, code, tables and leftover Markdown. It needs an NVIDIA GPU with 2.5 GiB VRAM, works standalone with any TTS system, is English-only for now, and a CPU-friendly version is on the way. Model and code are open on Hugging Face and GitHub.

Original post →

More from coding & agent

coding & agent channel →