audio/safeear/Dockerfile

FROM nvidia/cuda:12.1.1-cudnn8-runtime-ubuntu22.04

ENV DEBIAN_FRONTEND=noninteractive
ENV PYTHONUNBUFFERED=1

RUN apt-get update && apt-get install -y --no-install-recommends \
    python3 python3-pip python3-dev \
    git ffmpeg libsndfile1 wget \
    build-essential \
    && rm -rf /var/lib/apt/lists/*

RUN ln -sf /usr/bin/python3 /usr/bin/python

WORKDIR /app

# Install PyTorch with CUDA 12.1
RUN pip install --no-cache-dir \
    torch==2.5.1 torchaudio==2.5.1 \
    --index-url https://download.pytorch.org/whl/cu121

COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

# Clone SafeEar repository (for model code imports)
RUN git clone --depth 1 https://github.com/LetterLiGo/SafeEar.git /app/safeear_repo

# Install the fairseq fork with C extensions PATCHED OUT
# (same proven patch used by ShiftySpeech and Nes2Net --
#  C extensions are not needed for checkpoint loading)
WORKDIR /app/safeear_repo/fairseq_ours
RUN sed -i 's/ext_modules=extensions/ext_modules=[]/' setup.py && \
    pip install --no-cache-dir --no-deps -e .
WORKDIR /app

# Download model weights from HuggingFace
RUN mkdir -p /app/weights && \
    wget -q -O /app/weights/SpeechTokenizer.pt \
      "https://huggingface.co/TEC2004/SafeEar-ASV19-spoof-detection/resolve/main/SpeechTokenizer.pt" && \
    wget -q -O /app/weights/model.ckpt \
      "https://huggingface.co/TEC2004/SafeEar-ASV19-spoof-detection/resolve/main/model.ckpt"

COPY api.py .

EXPOSE 8002

RUN adduser --disabled-password --gecos '' appuser
USER appuser

CMD ["python", "api.py"]