Add vLLM audio image for transcription models

deploy/ai-server/docker-compose.audio.yml

@@ -1,6 +1,9 @@
 services:
   qwen-audio:
-    image: vllm/vllm-openai:latest
+    build:
+      context: .
+      dockerfile: vllm-audio.Dockerfile
+    image: vllm-audio:local
     container_name: qwen-audio
     profiles:
       - qwen-audio
@@ -53,7 +56,10 @@ services:
       start_period: 900s
 
   voxtral-small:
-    image: vllm/vllm-openai:latest
+    build:
+      context: .
+      dockerfile: vllm-audio.Dockerfile
+    image: vllm-audio:local
     container_name: voxtral-small
     profiles:
       - voxtral-small

deploy/ai-server/vllm-audio.Dockerfile

@@ -0,0 +1,3 @@
+FROM vllm/vllm-openai:latest
+
+RUN python3 -m pip install --no-cache-dir av soundfile librosa