8aa42e6fcb
resemblyzer depends on webrtcvad which compiles C extensions — gcc and build-essential are required in slim containers. silero-vad pip package has a broken native .so in python:3.12-slim; VAD model is now loaded at runtime via torch.hub.load() which downloads on first use. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
23 lines
981 B
Docker
23 lines
981 B
Docker
FROM python:3.12-slim
|
|
|
|
RUN apt-get update && apt-get install -y --no-install-recommends \
|
|
ffmpeg libsndfile1 gcc build-essential && rm -rf /var/lib/apt/lists/*
|
|
|
|
WORKDIR /app
|
|
|
|
COPY pyproject.toml .
|
|
# Install torch CPU-only first (avoids pulling GPU variant), then rest of deps
|
|
# silero-vad pip package excluded — runtime uses torch.hub.load instead
|
|
RUN pip install --no-cache-dir torch --index-url https://download.pytorch.org/whl/cpu && \
|
|
pip install --no-cache-dir resemblyzer scipy numpy && \
|
|
pip install --no-cache-dir fastapi uvicorn[standard] pydantic-settings httpx websockets livekit-api
|
|
|
|
COPY app/ app/
|
|
|
|
# Pre-download resemblyzer model (VoiceEncoder downloads ~25MB pretrained weights)
|
|
RUN python -c "from resemblyzer import VoiceEncoder; VoiceEncoder()"
|
|
# Note: silero-vad model downloads lazily at first container startup via torch.hub
|
|
|
|
ENV PYTHONUNBUFFERED=1
|
|
CMD ["python", "-m", "uvicorn", "app.pipeline.main:app", "--host", "0.0.0.0", "--port", "8300"]
|