You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

melo_mai_saya

ํ•œ๊ตญ์–ด TTS์šฉ์œผ๋กœ ํ•™์Šตํ•œ MeloTTS ๊ธฐ๋ฐ˜ ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค.

๊ธฐ์กด MeloTTS ํŒŒ์ดํ”„๋ผ์ธ์„ ์ปค์Šคํ„ฐ๋งˆ์ด์ฆˆํ•˜๊ณ  onnx ๋ณ€ํ™˜์„ ์ถ”๊ฐ€ํ–ˆ์Šต๋‹ˆ๋‹ค.
ํ˜„์žฌ ๋ฆฌํฌ์ง€ํ† ๋ฆฌ์—๋Š” ์„ธ ๊ฐ€์ง€ ํ™”์ž ๋ชจ๋ธ์ด ์žˆ์œผ๋ฉฐ,
๊ทธ์ค‘ ํ•˜๋‚˜๋Š” ๋ฐ์ดํ„ฐ์…‹์œผ๋กœ simon3000/genshin-voice๋ฅผ ์‚ฌ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค.
๋‚˜๋จธ์ง€ ๋‘ ๊ฐœ๋Š” voice conversion์œผ๋กœ ๋งŒ๋“  ๊ณ ์œ  ๋ฐ์ดํ„ฐ์…‹์œผ๋กœ ํ›ˆ๋ จ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.

์ž์„ธํ•œ ์‚ฌ์šฉ๋ฒ•์€ ๋ฆฌํฌ์ง€ํ† ๋ฆฌ๋ฅผ ์ฐธ๊ณ ํ•˜์„ธ์š”.

Sample

voice conversion์œผ๋กœ ๋งŒ๋“  ๋ชจ๋ธ์˜ ์ƒ˜ํ”Œ์ž…๋‹ˆ๋‹ค.

Model Details

  • Base repository: myshell-ai/MeloTTS
  • Language: Korean (ko)
  • BERT frontend: {BERT_MODEL_ID} (์˜ˆ: kykim/bert-kor-base)
  • Author: ์•ˆํ˜ธ์„ฑ (GitHub: hobi2k, Hugging Face: ahnhs2k)
  • Project repo: https://github.com/hobi2k/Kor_Voice_Lab

Training Data

๋ฐ์ดํ„ฐ๋Š” ์•„๋ž˜ ํ˜•์‹์˜ filelist/metadata๋ฅผ ์‚ฌ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค.

wav_path|speaker_name|language_code|text

์ „์ฒ˜๋ฆฌ ์ •์ฑ… ์˜ˆ์‹œ:

  • ํ•œ๊ตญ์–ด ํ…์ŠคํŠธ ํ•„ํ„ฐ๋ง
  • ํŠน์ˆ˜ ๊ธฐํ˜ธ/๋ฉ”ํƒ€ ํƒœ๊ทธ ์ œ๊ฑฐ
  • ์˜ค๋””์˜ค ๋ฆฌ์ƒ˜ํ”Œ๋ง ๋ฐ ํฌ๋งท ์ •๊ทœํ™”

Training Procedure

  • Framework: PyTorch
  • Training entry: torchrun -m melo.train
  • ์ฃผ์š” ์„ค์ •: {CONFIG_PATH_OR_SUMMARY}
  • Checkpoints:
    • G_*.pth (generator)
    • D_*.pth (discriminator)
    • DUR_*.pth (duration discriminator)

ONNX Export (Optional)

Kor_Voice_Lab์—์„œ๋Š” ONNX ๋ชจ๋ธ์„ ๋ณ„๋„๋กœ ์šด์˜ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

  • BERT ONNX
  • TTS ONNX

๊ด€๋ จ ์Šคํฌ๋ฆฝํŠธ:

  • scripts/bert_onnx_converter.py
  • scripts/onnx_converter.py
  • scripts/infer_onnx.py
  • tts_runtime/infer_onnx.py

Inference

PyTorch ์ถ”๋ก 

uv run python -m melo.infer \
  -t "์•ˆ๋…•ํ•˜์„ธ์š”. ํ…Œ์ŠคํŠธ ๋ฌธ์žฅ์ž…๋‹ˆ๋‹ค." \
  -m {G_CHECKPOINT_PATH} \
  -l KR \
  -o {OUT_DIR}

ONNX ์ถ”๋ก 

uv run python scripts/infer_onnx.py \
  --onnx {TTS_ONNX_PATH} \
  --bert {BERT_ONNX_PATH} \
  --config {CONFIG_PATH} \
  --text "์•ˆ๋…•ํ•˜์„ธ์š”. ONNX ์ถ”๋ก  ํ…Œ์ŠคํŠธ์ž…๋‹ˆ๋‹ค." \
  --speaker 0 \
  --lang KR \
  --device cpu \
  --out out.wav

Limitations

  • ๋ฐ์ดํ„ฐ์…‹ ๋„๋ฉ”์ธ/ํ™”์ž ํŽธํ–ฅ์ด ์Œ์ƒ‰๊ณผ ๋ฐœ์Œ ํ’ˆ์งˆ์— ์˜ํ–ฅ์„ ์ค„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  • ์ˆซ์ž/์™ธ๋ž˜์–ด/ํŠน์ˆ˜๊ธฐํ˜ธ๊ฐ€ ๋งŽ์€ ๋ฌธ์žฅ์—์„œ ๋ฐœ์Œ ์˜ค๋ฅ˜๊ฐ€ ๋ฐœ์ƒํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  • ONNX ํ’ˆ์งˆ์€ ๋ณ€ํ™˜ ์˜ต์…˜(opset, dynamic shape, ์ •๋ฐ€๋„)์— ๋ฏผ๊ฐํ•ฉ๋‹ˆ๋‹ค.

Citation

@misc{kor_voice_lab,
  title        = {Kor Voice Lab: Korean ASR-TTS Training, ONNX Conversion, and Runtime Pipeline},
  author       = {์•ˆํ˜ธ์„ฑ (GitHub: hobi2k)},
  year         = {2026},
  url          = {https://github.com/hobi2k/Kor_Voice_Lab},
  note         = {MeloTTS ๊ธฐ๋ฐ˜ ํ•œ๊ตญ์–ด ํ™•์žฅ ๋ฐ ONNX ๋Ÿฐํƒ€์ž„ ๊ฐœ์„ }
}

Acknowledgements


Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for ahnhs2k/melo_mai_saya

Quantized
(2)
this model

Dataset used to train ahnhs2k/melo_mai_saya

Collection including ahnhs2k/melo_mai_saya