Dockerfile

# OmniVoice TTS — HuggingFace Space (Docker SDK), T4 / dedicated GPU.
#
# Base: python:3.11-slim. We install torch 2.8.0 + cu128 ourselves to match
# the upstream k2-fsa/OmniVoice Space exactly. Using a generic Python base
# lets us upgrade torch without fighting a frozen CUDA layer and means the
# image works on either CPU or GPU hardware (torch's cu128 wheels statically
# link CUDA at runtime when a driver is present).

FROM python:3.11-slim

ENV PYTHONUNBUFFERED=1 \
    PIP_NO_CACHE_DIR=1 \
    PIP_DISABLE_PIP_VERSION_CHECK=1 \
    HF_HOME=/data/.cache/huggingface \
    HF_HUB_ENABLE_HF_TRANSFER=1 \
    TORCH_HOME=/data/.cache/torch \
    PORT=7860

# System deps for soundfile/librosa + audio + curl for HEALTHCHECK
RUN apt-get update && apt-get install -y --no-install-recommends \
        libsndfile1 ffmpeg git curl ca-certificates \
    && rm -rf /var/lib/apt/lists/*

WORKDIR /app

# Install torch first so other deps don't pull a different cuda variant.
# Versions match upstream k2-fsa/OmniVoice's requirements.txt.
RUN pip install --upgrade pip && \
    pip install torch==2.8.0 torchaudio==2.8.0 \
        --extra-index-url https://download.pytorch.org/whl/cu128

COPY requirements.txt /app/requirements.txt
RUN pip install -r /app/requirements.txt

COPY app.py /app/app.py

# HF Spaces map persistent state into /data; pre-create cache dirs so the model
# download lands in a writeable spot regardless of the runtime's UID.
RUN mkdir -p /data/.cache/huggingface /data/.cache/torch && chmod -R 777 /data

EXPOSE 7860

HEALTHCHECK --interval=30s --timeout=10s --retries=5 \
  CMD curl -fsS http://localhost:7860/health || exit 1

CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860", "--log-level", "info"]