Trindade7/TrdHiggsV3

Local TTS Voice Cloning for Non-English Languages — a Custom ComfyUI Node (with focus on European Portuguese / PT-PT & Angola) 🌐 Ler em: English | Português # Local TTS Voice Cloning for Non-English Languages — a Custom ComfyUI Node (with focus on European Portuguese / PT-PT & Angola) **Local European Portuguese TTS with voice cloning — a custom ComfyUI node** ## Background I recently moved to Portugal, and while waiting on my visa paperwork to clear (finally had my AIMA interview — all set now), I picked up some freelance work to keep busy (side note: if you've got an interesting AI, audio, or ComfyUI project, I'm currently available for hire — feel free to reach out! I'm a fullstack developer with solid experience integrating AI into production projects). One project stood out: a client wanted to convert text to speech using his own voice, entirely locally — no cloud uploads, for privacy reasons. Sounded straightforward at first, but he's a European Portuguese speaker, and it turned out most TTS models are trained heavily on Brazilian Portuguese. The accent, rhythm, and cadence just didn't match what he needed. To solve this properly, I ended up writing this custom ComfyUI node. It handles European Portuguese reliably, and it also chunks long text intelligently to avoid the quality degradation you typically get when feeding TTS models large blocks of text at once. I think this approach — and honestly the node itself — could be useful for anyone working with non-English languages where "the model technically supports it" doesn't mean it actually sounds right. ## Why Most TTS Pipelines Struggle with European Portuguese Most open-source TTS models advertise "Portuguese support," but in practice that means Brazilian Portuguese. European Portuguese has distinct vowel reduction, different prosody, and a cadence that Brazilian-trained models consistently get wrong. The result is output that sounds foreign to a native PT-PT listener. Higgs Audio v3 handles this significantly better, especially when paired with a voice cloning reference recorded by a native European Portuguese speaker. T …