Add private Vevo2 voice studio mode

This commit is contained in:
Mikei386
2026-09-09 13:38:00 +02:00
parent 535bd751b5
commit 68d02f32bd
15 changed files with 592 additions and 28 deletions
@@ -0,0 +1,64 @@
FROM mike-ai/bs-roformer-separator:0.47.0
ARG AMPHION_COMMIT=26f6883110181f1dbfe95c70a7c7dbaf4de5f42a
USER root
RUN apt-get update \
&& apt-get install -y --no-install-recommends git espeak-ng \
&& rm -rf /var/lib/apt/lists/*
RUN git clone https://github.com/open-mmlab/Amphion.git /opt/amphion \
&& cd /opt/amphion \
&& git checkout "${AMPHION_COMMIT}" \
&& rm -rf .git
# Vevo2 is tested on Athena with the CUDA/PyTorch stack inherited from the
# existing audio worker. Do not install Amphion's historical torch 2.0/cu118
# pins: Blackwell requires the newer cu128 runtime already present here.
RUN python -m pip install --no-cache-dir \
accelerate==1.10.1 \
diffusers==0.35.1 \
einops==0.8.1 \
easydict==1.13 \
g2p_en==2.1.0 \
humanfriendly==10.0 \
huggingface-hub==0.34.4 \
hydra-core==1.3.2 \
inflect==7.5.0 \
ipython==9.5.0 \
json5==0.12.1 \
librosa==0.11.0 \
loguru==0.7.3 \
matplotlib==3.10.6 \
munch==4.0.0 \
omegaconf==2.3.0 \
openai-whisper==20250625 \
phonemizer==3.3.0 \
python-multipart==0.0.20 \
praat-parselmouth==0.4.6 \
pypinyin==0.55.0 \
pyworld==0.3.5 \
ruamel.yaml==0.18.15 \
safetensors==0.6.2 \
tabulate==0.9.0 \
tgt==1.5 \
torchcrepe==0.0.24 \
transformers==4.56.1 \
typeguard==4.4.4 \
unidecode==1.4.0 \
vector-quantize-pytorch==1.12.5 \
vocos==0.1.0
WORKDIR /opt/amphion
ENV PYTHONPATH=/opt/amphion \
HF_HOME=/models/huggingface \
PYTHONUNBUFFERED=1
COPY run_fm_test.py /usr/local/bin/run_fm_test.py
COPY app.py /app/app.py
COPY index.html /app/index.html
EXPOSE 8008
HEALTHCHECK --interval=5s --timeout=3s --start-period=90s --retries=3 \
CMD curl -fsS http://127.0.0.1:8008/health || exit 1
ENTRYPOINT ["python", "/app/app.py"]