WordVoice TTS

#9 en Synthèse vocale (TTS)

Huggingapps · 2× · vu le 25 juil. 2026

29

Momentum

WordVoice TTS is a compact Text-to-Speech system (0.5 billion parameters) based on CosyVoice3, published by HuggingApps via a Hugging Face Space and the model "XXH333/WordVoice-base-0.5B". The model enables either fully automatic speech generation ("Autopilot") or explicit word-by-word control of duration, volume, pitch, and tone. It supports both cloned and preset voices. The underlying research paper "WordVoice" (arXiv 2607.06461) describes a framework for explicit, word-level control in LLM-based TTS systems, trained on a 4,700-hour bilingual dataset, among others.

Preuves (2)