Soften vocoder edge trim and shorter decode fades.

Align phoneme pipeline with modular reference; default UI lang en.

Tune Renikud Hebrew chunking to reduce stitched repetition.

Chunk Hebrew for Renikud only; single BlueTTS pass at chunk_len 150.

Pre-split Hebrew input before phonemization to stabilize long spans.

Move punctuation cleanup before phonemization

Normalize repeated punctuation before phonemization.

Fix flow matching loop and chunking logic

Fix 429 rate-limit: use hf_hub_download instead of raw requests

Auto RTL/LTR textarea: dir=auto + unicode-bidi:plaintext

Textarea: RTL for Hebrew via unicode-bidi plaintext and dir=auto

Document Space URL, upstream duplicate, link notmax123/blue

Add GitHub repo link in Space header

README title tweak; update male1 voice style

Use two voices only (Female/Male); drop female2 and male2

Update male1 voice style

chore: trigger Space rebuild

Update app.py and voice styles (female2, male1)

Sync app.py and gitignore; ignore .DS_Store

fix(build): drop lightblue-onnx git dep; add onnxruntime (Space uses app.py only)

Default language en: LANG_ID order, API defaults, tokenizer base_lang

docs/deps: point upstream to github.com/maxmelichov/BlueTTS

docs: refresh Space README for multilingual BlueTTS and four voices

Add four voices: Female 1/2, Male 1/2; rotate examples across voices

Update Gradio examples: same quote in Hebrew, EN, ES, IT, DE

Update female1 voice style

gitignore: block .env files so HF tokens are not committed

Sync: app updates, remove obsolete files

Ship new female voice style (female1.json); drop missing male1

Trigger Space rebuild

Trigger re-upload / rebuild

Fix: add missing synthesize() method to BlueTTS class

Improve controls row alignment: equal columns, baseline-aligned

Fix Gradio 6 compat: move css/theme to launch(), drop elem_classes from Examples

Use EuclidCircularB font embedded from fonts/EuclidCircularB.woff2

Clean release: multi-language BlueTTS with improved UI