feat: OuteTTS 엔진 추가(개인용/비상업 CC-BY-NC, 한국어)

- Qwen2 500M + WavTokenizer, torch cu128 GPU, 전용 venv 워커(8003)
- 한국어 프리셋 4종(male/female), 속도·피치 librosa 후처리
- transformers 4.49로 torchvision 강제 import 회피, soundfile 저장
- scenema-audio는 VRAM 24GB+/Gemma12B 게이트로 8GB GPU에선 불가 → 미포함
This commit is contained in:
claude
2026-08-25 10:02:24 +09:00
parent ed09c6a08f
commit 1aa0bbf3e6
6 changed files with 230 additions and 1 deletions

View File

@@ -72,11 +72,25 @@ RUN HF_HOME=/models /opt/supertonic-venv/bin/python -c "from supertonic import T
ENV SUPERTONIC_URL=http://127.0.0.1:8002
# ===== OuteTTS 엔진 (개인용/비상업, CC-BY-NC-4.0) =====
# Qwen2 500M LM + WavTokenizer 코덱. torch cu128 GPU. transformers 4.49(토치비전 회피).
COPY outetts_worker /opt/outetts_worker
RUN python -m venv /opt/oute-venv \
&& /opt/oute-venv/bin/pip install --no-cache-dir torch==2.11.0 torchaudio==2.11.0 \
--index-url https://download.pytorch.org/whl/cu128 \
&& /opt/oute-venv/bin/pip install --no-cache-dir -r /opt/outetts_worker/requirements.txt
RUN /opt/oute-venv/bin/pip uninstall -y torchvision 2>/dev/null || true
# 모델+코덱 사전 다운로드(베이킹) — 빌드 시엔 GPU 없이 CPU 로 로드
RUN HF_HOME=/models /opt/oute-venv/bin/python -c "import outetts; cfg=outetts.HFModelConfig_v1(model_path='OuteAI/OuteTTS-0.2-500M', language='ko'); outetts.InterfaceHF(model_version='0.2', cfg=cfg); print('oute warmup ok')"
ENV OUTETTS_URL=http://127.0.0.1:8003
COPY app ./app
COPY frontend ./frontend
COPY entrypoint.sh /entrypoint.sh
RUN chmod +x /entrypoint.sh
# MeloTTS(8788, 메인) + Supertonic 워커(8002) 동시 기동
# MeloTTS(8788) + Supertonic(8002) + OuteTTS(8003) 동시 기동
EXPOSE 8788
CMD ["/entrypoint.sh"]