First attempt at diarization (with custom Dockerfile extension).

whisper/.gitignore

@@ -0,0 +1 @@
+whisper-live.env

whisper/compose.yaml

@@ -1,23 +1,35 @@
 services:
   whisper-live:
-    image: ghcr.io/collabora/whisperlive-openvino:latest
+    image: whisper-live-diarization:latest
     container_name: whisper-live
     restart: unless-stopped
     ports:
-      - "9097:9090"   # Maps your existing client port 9097 to the official 9090 port
+      - "9097:9090"
     volumes:
       - whisper-live-data:/var/lib/whisper-live
-    
-    # Tells WhisperLive to use the OpenVINO hardware engine
-    command: ["python3", "run_server.py", "--port", "9090", "--backend", "openvino", "--omp_num_threads", "4", "--max_clients", "1", "--max_connection_time", "7200", "--raw_pcm_input"]
-
+      - whisper-pyannote-cache:/root/.cache/huggingface/hub
+    env_file:
+      - whisper-live.env          # contains HUGGING_FACE_HUB_TOKEN — not in git
+    command:
+      - "python3"
+      - "run_server.py"
+      - "--port"
+      - "9090"
+      - "--backend"
+      - "openvino"
+      - "--omp_num_threads"
+      - "4"
+      - "--max_clients"
+      - "1"
+      - "--max_connection_time"
+      - "7200"
+      - "--raw_pcm_input"
     devices:
       - /dev/dri/renderD128:/dev/dri/renderD128
-      - /dev/dri/card1:/dev/dri/card1  # Change to card0 if your 13700K is the only GPU
-
-    cpuset: "0-7"          # 4 P-cores / 8 logical threads — sufficient for OpenVINO orchestration
-    mem_limit: 10g         # confirmed adequate from OOM testing
+      - /dev/dri/card1:/dev/dri/card1
+    cpuset: "0-7"
+    mem_limit: 10g
 
 volumes:
   whisper-live-data:
-
+  whisper-pyannote-cache: