Replace XTTS with Qwen3-TTS

compose.yaml

@@ -722,9 +722,9 @@ services:
       retries: 30
       start_period: 120s
 
-  xtts:
-    image: ${XTTS_IMAGE:-ghcr.io/coqui-ai/xtts-streaming-server:latest-cuda121@sha256:f7fb3b1f9d4bc88af94da1b5959d8002f1e0b003c97557164034eb8a29f01b90}
-    container_name: mike-ai-xtts
+  qwen3-tts:
+    image: ${QWEN3_TTS_IMAGE:-ghcr.io/malaiwah/qwen3-tts-server:latest@sha256:b363a01d08b1bbecbfc3ca6f585368fae2cfdc591f9ecca6643738369f9a9d98}
+    container_name: mike-ai-qwen3-tts
     restart: unless-stopped
     deploy:
       resources:
@@ -732,30 +732,33 @@ services:
           devices:
             - driver: nvidia
               device_ids:
-                - ${XTTS_GPU_DEVICE:-GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b}
+                - ${QWEN3_TTS_GPU_DEVICE:-GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b}
               capabilities: [gpu]
     read_only: true
     shm_size: 1g
     tmpfs:
       - /tmp:size=1g,mode=1777
-      - /root/.cache:size=2g,mode=0700
     volumes:
-      - "${XTTS_CACHE_DIR:-/data/models/xtts-v2-cache}:/root/.local/share/tts"
+      - "${QWEN3_TTS_CACHE_DIR:-/data/models/qwen3-tts-cache}:/root/.cache/huggingface"
+      - "${QWEN3_TTS_VOICES_DIR:-/data/models/qwen3-tts-voices}:/data/voices"
     environment:
-      COQUI_TOS_AGREED: "1"
-      NVIDIA_VISIBLE_DEVICES: ${XTTS_GPU_DEVICE:-GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b}
+      NVIDIA_VISIBLE_DEVICES: ${QWEN3_TTS_GPU_DEVICE:-GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b}
       NVIDIA_DRIVER_CAPABILITIES: compute,utility
       CUDA_VISIBLE_DEVICES: "0"
-      NUM_THREADS: "4"
+      HF_HOME: /root/.cache/huggingface
+      NUMBA_CACHE_DIR: /tmp/numba
+      QWEN3_TTS_MODEL_ID: Qwen/Qwen3-TTS-12Hz-1.7B-Base
+      QWEN3_TTS_DEFAULT_VOICE: serena
+      QWEN3_TTS_VOICES_DIR: /data/voices
     networks: [frontend]
     security_opt: ["no-new-privileges:true"]
     cap_drop: [ALL]
     healthcheck:
-      test: [CMD, curl, -fsS, "http://127.0.0.1/languages"]
+      test: [CMD, python, -c, "import urllib.request; urllib.request.urlopen('http://127.0.0.1:8001/health', timeout=2)"]
       interval: 10s
       timeout: 5s
-      retries: 36
-      start_period: 240s
+      retries: 60
+      start_period: 600s
 
   tts-gateway:
     build:
@@ -769,22 +772,17 @@ services:
     environment:
       TTS_GATEWAY_HOST: 0.0.0.0
       TTS_GATEWAY_PORT: "8085"
-      XTTS_URL: http://xtts:80
+      QWEN_TTS_URL: http://qwen3-tts:8001
+      QWEN_TTS_MODEL: tts-1
+      QWEN_TTS_VOICE: serena
+      QWEN_TTS_LANGUAGE: German
+      QWEN_TTS_TIMEOUT: "120"
       PIPER_URL: http://piper:8085
       TTS_VOICE_ALIAS: alloy
-      XTTS_SPEAKER: Annmarie Nele
       TTS_DEFAULT_LANGUAGE: de
       # Mixed-language clip stitching caused long pauses and unintelligible
       # transitions. Keep full sentences in one stable German voice.
       TTS_CODE_SWITCH_ENABLED: "false"
-      XTTS_QUEUE_TIMEOUT: "15"
-      XTTS_TIMEOUT: "120"
-      # Keep normal sentences intact for natural prosody. This is only the
-      # safety ceiling for unusually long sentences.
-      XTTS_CHUNK_CHARS: "420"
-      # XTTS occasionally inserts multi-second silence inside a phrase.
-      XTTS_INTERNAL_SILENCE_TRIGGER_MS: "650"
-      XTTS_INTERNAL_SILENCE_KEEP_MS: "220"
       PIPER_TIMEOUT: "120"
     networks: [frontend]
     depends_on: