Update Dockerfile

Dockerfile

@@ -47,6 +47,15 @@ RUN python -c "from huggingface_hub import snapshot_download; \
     snapshot_download('hubertsiuzdak/snac_24khz')" \
     || echo 'WARNING: SNAC pre-download failed'
 
+# Pre-download FLOAT dependencies (wav2vec2 + emotion model) — cached in Docker layer
+RUN mkdir -p /app/checkpoints && \
+    python -c "from huggingface_hub import snapshot_download; \
+    print('Downloading wav2vec2-base-960h...'); \
+    snapshot_download('facebook/wav2vec2-base-960h', local_dir='/app/checkpoints/wav2vec2-base-960h'); \
+    print('Downloading emotion model...'); \
+    snapshot_download('r-f/wav2vec-english-speech-emotion-recognition', local_dir='/app/checkpoints/wav2vec-english-speech-emotion-recognition'); \
+    print('Models ready')"
+
 # Download Vosk small English model (~50MB, CPU-only STT)
 RUN python -c "import urllib.request, zipfile, os; \
     url='https://alphacephei.com/vosk/models/vosk-model-small-en-us-0.15.zip'; \
@@ -64,6 +73,9 @@ COPY --chown=user ./app.py ./llm.py ./groq_tts.py ./orpheus_tts.py ./float_lipsy
 # Copy avatar profiles (each has ref.png, persona.txt, idlevideos/)
 COPY --chown=user ./avatars/ /app/avatars/
 
+# Copy checkpoints (wav2vec2, emotion model, float.pth) — baked in to avoid runtime downloads
+COPY --chown=user ./app/checkpoints/ /app/checkpoints/
+
 # Copy voice reference + transcript for voice cloning (used by orpheus_tts)
 COPY --chown=user ./voice* /app/