fix: pad with X (id 0), the token training used, not the word separator

fix: pad with X (id 0), the token training used, not the word separator

docs: document the BSC per-speaker style blobs (11 ca + 6 es named speakers)

Upload folder using huggingface_hub

Upload folder using huggingface_hub

card: ProxectoNos Galician StyleTTS2 voices

add proxectonos-gl-brais (ProxectoNos Galician StyleTTS2)

add proxectonos-gl-celtia (ProxectoNos Galician StyleTTS2)

docs: add model card

Upload bsc-ca-styletts2/style_encoder.onnx with huggingface_hub

Upload bsc-ca-styletts2/config.json with huggingface_hub

Upload bsc-ca-styletts2/model.onnx with huggingface_hub

Upload bsc-es-styletts2/style_encoder.onnx with huggingface_hub

Upload bsc-es-styletts2/config.json with huggingface_hub

Upload bsc-es-styletts2/model.onnx with huggingface_hub

Add HiTZ/StyleTTS2-eu_emo: emotional Basque (antton/maider × neu/poz/tri)

Add HiTZ StyleTTS2-eu (Basque, antton+maider styles) — onnx+config+styles

lang_code -> ja-JP (BCP-47)

zh -> misaki_zh + alphabet=bopomofo

zh -> misaki_zh_v11 (no version param)

phoneme_type -> misaki_en

phoneme_type -> misaki_zh

phoneme_type -> misaki_en

phoneme_type -> misaki_zh

phoneme_type -> misaki_ja

phoneme_type -> misaki_en

Kokoro v0.19 legacy int8

Kokoro hi espeak config

Kokoro es espeak config

Kokoro fr espeak config

Kokoro pt espeak config

Kokoro it espeak config

Kokoro v1.1-zh int8 + zh(1.1)/en configs

Kokoro zh config

Kokoro ja config

Kokoro-82M fp16 + misaki config

pure StyleTTS2 (stitched DDATT pipeline)

initial commit