Remove Beta 1 and Piper fallback

2026-09-10 13:17:57 +02:00
parent 52482627be
commit 08ff3d7c4e
38 changed files with 94 additions and 484 deletions
@@ -6,8 +6,6 @@ CONTROLLER_TOKEN=GENERATED_BY_INSTALLER
 FLUX_COMPONENT_DIR=/data/models/FLUX.2-klein-9B-components
 FLUX_TRANSFORMER_DIR=/data/models/FLUX.2-klein-9B-fp8
 IMAGE_GPU_DEVICES=GPU-8ad38c6c-5a01-9d8e-1dfa-ed662ad78fbe
-PIPER_TTS_VERSION=1.6.0
-PIPER_VOICE=de_DE-thorsten-high
 QWEN3_TTS_IMAGE=ghcr.io/malaiwah/qwen3-tts-server:latest@sha256:b363a01d08b1bbecbfc3ca6f585368fae2cfdc591f9ecca6643738369f9a9d98
 QWEN3_TTS_CACHE_DIR=/data/models/qwen3-tts-cache
 QWEN3_TTS_VOICES_DIR=/data/models/qwen3-tts-voices
@@ -17,7 +15,6 @@ DEFAULT_REASONING_EFFORT=off

 FAST_MODEL_FILE=qwen-mix/Qwen3.8-27B-IQ4-MIX.gguf
 MEDIUM_MODEL_FILE=qwen-pure/qwen3.8-27b-IQ4_XS-pure.gguf
-BETA1_MODEL_FILE=qwen3.8-27b-gsq-rco-iq3s/Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp.gguf
 LARGE_MODEL_FILE=qwen-pure/qwen3.8-27b-IQ4_XS-pure.gguf
 ULTRA_MODEL_FILE=qwen-pure/qwen3.8-27b-IQ4_XS-pure.gguf
 UNCENSORED_MODEL_FILE=qwen3.8-27b-abliterated/Qwen3.8-27B-ABLITERATED-Q4_K_M.gguf
@@ -30,9 +27,6 @@ FAST_UBATCH_SIZE=32
 MEDIUM_CONTEXT=160000
 MEDIUM_BATCH_SIZE=2048
 MEDIUM_UBATCH_SIZE=128
-BETA1_CONTEXT=112000
-BETA1_BATCH_SIZE=2048
-BETA1_UBATCH_SIZE=128
 LARGE_CONTEXT=192000
 LARGE_BATCH_SIZE=2048
 LARGE_UBATCH_SIZE=128
@@ -45,7 +39,6 @@ UNCENSORED_UBATCH_SIZE=128
 FAST_GPU_DEVICES=GPU-8ad38c6c-5a01-9d8e-1dfa-ed662ad78fbe,GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b
 MEDIUM_GPU_DEVICES=GPU-8ad38c6c-5a01-9d8e-1dfa-ed662ad78fbe,GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b
 MEDIUM_TENSOR_SPLIT=85,15
-BETA1_GPU_DEVICES=GPU-8ad38c6c-5a01-9d8e-1dfa-ed662ad78fbe,GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b
 LARGE_GPU_DEVICES=GPU-8ad38c6c-5a01-9d8e-1dfa-ed662ad78fbe,GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b
 LARGE_TENSOR_SPLIT=86,14
 ULTRA_GPU_DEVICES=GPU-8ad38c6c-5a01-9d8e-1dfa-ed662ad78fbe,GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b
@@ -58,7 +51,6 @@ LLAMA_THREADS_BATCH=6
 FAST_PARALLEL_SLOTS=1
 LLAMA_CACHE_RAM_MIB=32768
 MEDIUM_PARALLEL_SLOTS=1
-BETA1_PARALLEL_SLOTS=1
 LARGE_PARALLEL_SLOTS=1
 ULTRA_PARALLEL_SLOTS=1
 UNCENSORED_PARALLEL_SLOTS=1
@@ -11,7 +11,7 @@ Sie betreibt:
 - llama.cpp mit genau einem aktiven Qwen-Profil,
 - den OpenAI-kompatiblen Profile Router,
 - FLUX.2 Klein 9B FP8 Beta für Textbilder und Referenzbild-Bearbeitung,
- Qwen3-TTS und Piper für Sprache,
+- Qwen3-TTS für Sprache,
 - ACE-Step 1.5 XL-SFT als exklusiven Musikstudio-Modus,
 - das Athena-Dashboard,
 - Portainer CE als optionale Ansicht auf die laufenden Docker-Container,
@@ -66,7 +66,7 @@ Qwen-Profil wird vom Profile Controller verwaltet.
  llama.cpp-Profil und Qwen3-TTS werden dafür gestoppt und danach automatisch
  wiederhergestellt. Die Beta arbeitet mit 1024 × 1024 Pixeln, vier Schritten
  und Guidance 1,0.
- Qwen3-TTS 1.7B: RTX 3060; Piper bleibt CPU-Fallback. Der Router reicht
+- Qwen3-TTS 1.7B: RTX 3060. Der Router reicht
  zusätzlich natives 24-kHz-PCM für den optionalen Hermes-Streaming-Adapter
  unter `integrations/hermes-qwen3-stream` durch.
 - ACE-Step 1.5 XL-SFT: exklusiver Musikmodus auf der RTX 5080. Dashboard und
@@ -12,7 +12,7 @@ Bild- und Sprachausgabe. **Hermes und die Fach-MCPs laufen auf Unraid.**
 - Profile Router auf Port 8081
 - FLUX.2 Klein 9B FP8 Beta für Textbilder und Referenzbild-Bearbeitung:
  Transformer auf RTX 5080, Qwen3-8B-NF4-Textencoder auf RTX 3060
- Qwen3-TTS 1.7B auf der RTX 3060 mit Piper als CPU-Fallback
+- Qwen3-TTS 1.7B auf der RTX 3060 hinter einem Normalisierungs- und Streaming-Gateway
 - Whisper.cpp `ggml-small` auf der CPU für lokale deutsche Spracherkennung
 - Live-Dashboard mit 21 Tagen Detailhistorie auf Port 8099
 - Dashboard-Umschaltung zwischen LLM-Betrieb, ACE-Step-Musikstudio,
@@ -110,8 +110,8 @@ zuerst das aktive llama.cpp-Profil und Qwen3-TTS. Anschließend läuft der
 FP8-Transformer auf der RTX 5080 und der in NF4 geladene Qwen3-8B-Textencoder
 auf der RTX 3060. Vor dem VAE-Decoding werden Transformer und Textencoder
 freigegeben. Nach dem Bildauftrag stoppt der Router den Bild-Worker und stellt
-Qwen3-TTS sowie das zuvor aktive Textprofil automatisch wieder her. Piper
-bleibt währenddessen als CPU-Fallback verfügbar.
+Qwen3-TTS sowie das zuvor aktive Textprofil automatisch wieder her. Während
+der exklusiven Nutzung der RTX 3060 ist TTS vorübergehend nicht verfügbar.

 Die Beta ist derzeit bewusst auf `1024x1024`, vier Schritte, Guidance `1.0`,
 einen parallelen Auftrag und maximal vier lokale Referenzbilder begrenzt.
@@ -230,89 +230,6 @@ services:
      - --spec-draft-p-min
      - "0.05"

-  # Beta 1 keeps the complete GSQ-RCO IQ3_S text model and KV cache on the RTX
-  # 5080. Only the multimodal projector runs on the RTX 3060. The larger IQ3_S
-  # build starts conservatively at 112K until its hard context boundary has
-  # been measured; the former IQ3_XXS result does not transfer to this file.
-  llama-beta1:
-    <<: *llama-common
-    container_name: mike-ai-llama-beta1
-    labels:
-      com.mike-ai.llama-profile: beta1
-    environment:
-      NVIDIA_VISIBLE_DEVICES: ${BETA1_GPU_DEVICES:-0,1}
-      NVIDIA_DRIVER_CAPABILITIES: compute,utility
-      MTMD_BACKEND_DEVICE: CUDA1
-    command:
-      - --model
-      - "/models/${BETA1_MODEL_FILE:?BETA1_MODEL_FILE is required}"
-      - --mmproj
-      - "/models/${VISION_PROJECTOR_FILE:?VISION_PROJECTOR_FILE is required}"
-      - --mmproj-offload
-      - --mmproj-device
-      - CUDA1
-      - --alias
-      - qwen-beta-1
-      - --ctx-size
-      - "${BETA1_CONTEXT:-112000}"
-      - --flash-attn
-      - "on"
-      - --cache-type-k
-      - q4_0
-      - --cache-type-v
-      - q4_0
-      - --cache-prompt
-      - --cache-ram
-      - "${LLAMA_CACHE_RAM_MIB:-32768}"
-      - --threads
-      - "${LLAMA_THREADS:-6}"
-      - --threads-batch
-      - "${LLAMA_THREADS_BATCH:-6}"
-      - --batch-size
-      - "${BETA1_BATCH_SIZE:-2048}"
-      - --ubatch-size
-      - "${BETA1_UBATCH_SIZE:-128}"
-      - --parallel
-      - "${BETA1_PARALLEL_SLOTS:-1}"
-      - --kv-unified
-      - --jinja
-      - --reasoning
-      - auto
-      - --reasoning-preserve
-      - --host
-      - 0.0.0.0
-      - --port
-      - "8080"
-      - --metrics
-      - --fit
-      - "off"
-      - --n-gpu-layers
-      - all
-      - --no-mmap
-      - --no-ui
-      - --temperature
-      - "1.0"
-      - --top-p
-      - "0.95"
-      - --top-k
-      - "20"
-      - --device
-      - CUDA0
-      - --main-gpu
-      - "0"
-      - --split-mode
-      - none
-      - --spec-type
-      - draft-mtp
-      - --spec-draft-n-max
-      - "3"
-      - --spec-draft-type-k
-      - f16
-      - --spec-draft-type-v
-      - f16
-      - --spec-draft-p-min
-      - "0.05"
-
  llama-large:
    <<: *llama-common
    container_name: mike-ai-llama-large
@@ -565,7 +482,7 @@ services:
      - /var/run/docker.sock:/var/run/docker.sock
    environment:
      CONTROLLER_TOKEN: "${CONTROLLER_TOKEN:?CONTROLLER_TOKEN is required}"
-      ALLOWED_PROFILES: fast,medium,beta1,large,ultra,uncensored
+      ALLOWED_PROFILES: fast,medium,large,ultra,uncensored
      IMAGE_WORKER: image
      RESTORE_WORKER: restore
      TTS_WORKER: qwen3
@@ -622,11 +539,10 @@ services:
      CHAT_IMAGE_ALLOW_REMOTE_URLS: "false"
      ENABLE_IMAGE_GENERATION: "true"
      ENABLE_TTS: "true"
-      # Stable OpenAI compatibility names remain piper/alloy for existing
-      # clients. The gateway maps them to Qwen3-TTS and falls back to Piper
-      # while Qwen3-TTS is unavailable, busy or paused for image generation.
+      # The gateway keeps text normalization, output conversion and native
+      # PCM streaming in one stable API in front of Qwen3-TTS.
      TTS_WORKER_URL: http://tts-gateway:8085
-      TTS_MODEL: piper
+      TTS_MODEL: qwen3-tts
      TTS_VOICES: alloy
      TTS_DEFAULT_VOICE: alloy
      ENABLE_STT: "true"
@@ -655,8 +571,6 @@ services:
        condition: service_healthy
      profile-controller:
        condition: service_healthy
-      piper:
-        condition: service_healthy
      tts-gateway:
        condition: service_healthy
      whisper:
@@ -699,37 +613,6 @@ services:
      timeout: 3s
      retries: 12

-  piper:
-    build:
-      context: platform/docker/piper
-      args:
-        PIPER_TTS_VERSION: ${PIPER_TTS_VERSION:-1.6.0}
-    image: mike-ai/piper:local
-    container_name: mike-ai-piper
-    restart: unless-stopped
-    read_only: true
-    tmpfs:
-      - /tmp:size=256m,mode=1777
-    volumes:
-      - piper-data:/data
-    environment:
-      PIPER_DATA_DIR: /data
-      PIPER_VOICE: ${PIPER_VOICE:-de_DE-thorsten-high}
-      PIPER_VOICE_ALIAS: alloy
-      PIPER_HOST: 0.0.0.0
-      PIPER_PORT: "8085"
-      PIPER_MAX_TEXT_CHARS: "8000"
-    networks: [frontend]
-    security_opt: ["no-new-privileges:true"]
-    cap_drop: [ALL]
-    cap_add: [CHOWN, SETUID, SETGID]
-    healthcheck:
-      test: [CMD, curl, -fsS, "http://127.0.0.1:8085/status"]
-      interval: 10s
-      timeout: 5s
-      retries: 30
-      start_period: 120s
-
  qwen3-tts:
    image: ${QWEN3_TTS_IMAGE:-ghcr.io/malaiwah/qwen3-tts-server:latest@sha256:b363a01d08b1bbecbfc3ca6f585368fae2cfdc591f9ecca6643738369f9a9d98}
    container_name: mike-ai-qwen3-tts
@@ -787,17 +670,12 @@ services:
      QWEN_TTS_VOICE: serena
      QWEN_TTS_LANGUAGE: German
      QWEN_TTS_TIMEOUT: "120"
-      PIPER_URL: http://piper:8085
      TTS_VOICE_ALIAS: alloy
      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"
-      PIPER_TIMEOUT: "120"
    networks: [frontend]
-    depends_on:
-      piper:
-        condition: service_healthy
    security_opt: ["no-new-privileges:true"]
    cap_drop: [ALL]
    healthcheck:
@@ -920,7 +798,6 @@ services:
      - /data/docker-backups:/archive
      - /etc/mike-ai:/backup/etc-mike-ai:ro
      - /opt/mike-ai/stack:/backup/stack:ro
-      - piper-data:/backup/volumes/piper-data:ro
      - router-state:/backup/volumes/router-state:ro
      - router-images:/backup/volumes/router-images:ro
      - portainer-data:/backup/volumes/portainer-data:ro
@@ -947,7 +824,6 @@ networks:
    name: mike-ai-tools-egress

 volumes:
-  piper-data:
  whisper-data:
  router-state:
  router-images:
@@ -60,9 +60,6 @@ FAST_MODEL_SHA256=54879ae8738d5938f46cb3b8cbf16bf42b8c85b7d68d7c73f062b612ec183e
 MEDIUM_MODEL_FILE=qwen-pure/qwen3.8-27b-IQ4_XS-pure.gguf
 MEDIUM_MODEL_URL=https://huggingface.co/jpetrina/Qwen3.8-27B-IQ4_XS-pure-GGUF/resolve/main/qwen3.8-27b-IQ4_XS-pure.gguf
 MEDIUM_MODEL_SHA256=ea5a3c45d407f9b9e5d2c0d647f0ea600f486f6b86b92b56d0823ba073dae675
-BETA1_MODEL_FILE=qwen3.8-27b-gsq-rco-iq3s/Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp.gguf
-BETA1_MODEL_URL=https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF/resolve/main/Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp.gguf
-BETA1_MODEL_SHA256=58fd826723939933dc86f45b7fe04545cbc2de1c70f6fe2cdd3858c87a98c12f
 LARGE_MODEL_FILE=qwen-pure/qwen3.8-27b-IQ4_XS-pure.gguf
 LARGE_MODEL_URL=https://huggingface.co/jpetrina/Qwen3.8-27B-IQ4_XS-pure-GGUF/resolve/main/qwen3.8-27b-IQ4_XS-pure.gguf
 LARGE_MODEL_SHA256=ea5a3c45d407f9b9e5d2c0d647f0ea600f486f6b86b92b56d0823ba073dae675
@@ -89,11 +86,6 @@ MEDIUM_CONTEXT=160000
 MEDIUM_BATCH_SIZE=2048
 MEDIUM_UBATCH_SIZE=128
 MEDIUM_TENSOR_SPLIT=85,15
-BETA1_CONTEXT=112000
-BETA1_BATCH_SIZE=2048
-BETA1_UBATCH_SIZE=128
-BETA1_GPU_DEVICES=GPU-8ad38c6c-5a01-9d8e-1dfa-ed662ad78fbe,GPU-4834d9d7-5b61-3004-1fb3-4ae49d482d4b
-BETA1_PARALLEL_SLOTS=1
 LARGE_CONTEXT=192000
 LARGE_BATCH_SIZE=2048
 LARGE_UBATCH_SIZE=128
@@ -115,8 +107,6 @@ MEDIUM_PARALLEL_SLOTS=1
 LARGE_PARALLEL_SLOTS=1
 ULTRA_PARALLEL_SLOTS=1
 UNCENSORED_PARALLEL_SLOTS=1
-PIPER_TTS_VERSION=1.6.0
-PIPER_VOICE=de_DE-thorsten-high
 QWEN3_TTS_IMAGE=ghcr.io/malaiwah/qwen3-tts-server:latest@sha256:b363a01d08b1bbecbfc3ca6f585368fae2cfdc591f9ecca6643738369f9a9d98
 QWEN3_TTS_CACHE_DIR=/data/models/qwen3-tts-cache
 QWEN3_TTS_VOICES_DIR=/data/models/qwen3-tts-voices
@@ -27,18 +27,6 @@
      "mtp": 3,
      "description": "Ausgewogenes Standardprofil für Alltag und lange agentische Aufgaben."
    },
-    {
-      "id": "beta1",
-      "alias": "qwen-beta-1",
-      "context": 112000,
-      "parallel_slots": 1,
-      "model_env": "BETA1_MODEL_FILE",
-      "model_family": "Qwen3.8-27B GSQ-RCO IQ3_S MTP",
-      "gpu_split": "5080 model / 3060 vision",
-      "vision": true,
-      "mtp": 3,
-      "description": "Beta 1: qualitaetsoptimiertes GSQ-RCO-IQ3_S-Testprofil; 112K Startkontext bis zur erneuten Grenzmessung."
-    },
    {
      "id": "large",
      "alias": "qwen-large",
@@ -71,8 +71,8 @@ class Handler(BaseHTTPRequestHandler):
            self._send_json(200, {
                "status": "ok",
                "ready": True,
-                "voices": ["claribel"],
-                "default_voice": "claribel",
+                "voices": ["alloy"],
+                "default_voice": "alloy",
                "load_errors": [],
                "sample_rate": SAMPLE_RATE,
                "uptime_seconds": 1.0,
@@ -561,14 +561,14 @@ d=json.load(sys.stdin)
 tts=d["tts"]
 assert tts["reachable"] is True, tts
 assert tts["ready"] is True, tts
-assert set(tts["voices"])=={"claribel"}, tts
-' && ok "Status: TTS erreichbar, bereit, 2 Stimmen" || bad "Status tts-Section"
+assert set(tts["voices"])=={"alloy"}, tts
+' && ok "Status: TTS erreichbar und bereit" || bad "Status tts-Section"

 # --- 28. TTS: POST /v1/audio/speech (wav) ---------------------------------------------------------------
 echo "== Test 28: POST /v1/audio/speech (wav)"
 CODE=$(curl -s -o /tmp/tts28.wav -w "%{http_code}" -D /tmp/hdr28.txt \
  "$BASE/v1/audio/speech" -H "Content-Type: application/json" \
-  -d '{"model":"xtts-v2","input":"Hallo Welt","voice":"claribel","response_format":"wav"}')
+  -d '{"model":"qwen3-tts","input":"Hallo Welt","voice":"alloy","response_format":"wav"}')
 CTYPE=$(grep -i content-type /tmp/hdr28.txt | tr -d "\r")
 [ "$CODE" = "200" ] && [ -s /tmp/tts28.wav ] && echo "$CTYPE" | grep -qi "audio/wav" \
  && ok "TTS wav (200, $CTYPE, $(stat -f%z /tmp/tts28.wav 2>/dev/null || stat -c%s /tmp/tts28.wav) Bytes)" \
@@ -578,7 +578,7 @@ CTYPE=$(grep -i content-type /tmp/hdr28.txt | tr -d "\r")
 echo "== Test 29: POST /v1/audio/speech (mp3, Default)"
 CODE=$(curl -s -o /tmp/tts29.mp3 -w "%{http_code}" -D /tmp/hdr29.txt \
  "$BASE/v1/audio/speech" -H "Content-Type: application/json" \
-  -d '{"input":"Guten Tag","voice":"claribel"}')
+  -d '{"input":"Guten Tag","voice":"alloy"}')
 CTYPE=$(grep -i content-type /tmp/hdr29.txt | tr -d "\r")
 [ "$CODE" = "200" ] && [ -s /tmp/tts29.mp3 ] && echo "$CTYPE" | grep -qi "audio/mpeg" \
  && ok "TTS mp3 (200, $CTYPE)" || bad "TTS mp3 (Code $CODE, $CTYPE)"
@@ -586,7 +586,7 @@ CTYPE=$(grep -i content-type /tmp/hdr29.txt | tr -d "\r")
 # --- 30. TTS: Validierung --------------------------------------------------------------------------------
 echo "== Test 30: TTS-Validierung"
 CODE=$(curl -s -o /tmp/err30a.json -w "%{http_code}" "$BASE/v1/audio/speech" \
-  -H "Content-Type: application/json" -d '{"voice":"claribel"}')
+  -H "Content-Type: application/json" -d '{"voice":"alloy"}')
 cat /tmp/err30a.json; echo
 [ "$CODE" = "400" ] && ok "400 bei fehlendem input" || bad "erwartet 400, bekam $CODE"

@@ -608,7 +608,7 @@ cat /tmp/err30d.json; echo
 # --- 31. TTS: Worker-Fehler → 503 ------------------------------------------------------------------------
 echo "== Test 31: TTS-Worker-Fehler → 503"
 CODE=$(curl -s -o /tmp/err31.json -w "%{http_code}" "$BASE/v1/audio/speech" \
-  -H "Content-Type: application/json" -d '{"input":"FAIL","voice":"claribel"}')
+  -H "Content-Type: application/json" -d '{"input":"FAIL","voice":"alloy"}')
 cat /tmp/err31.json; echo
 [ "$CODE" = "503" ] && ok "503 bei TTS-Worker-Fehler" || bad "erwartet 503, bekam $CODE"

@@ -617,7 +617,7 @@ echo "== Test 32: TTS-Worker down → 503"
 kill "$TTS_PID" 2>/dev/null || true
 sleep 0.5
 CODE=$(curl -s -o /tmp/err32.json -w "%{http_code}" "$BASE/v1/audio/speech" \
-  -H "Content-Type: application/json" -d '{"input":"Hallo","voice":"claribel"}')
+  -H "Content-Type: application/json" -d '{"input":"Hallo","voice":"alloy"}')
 cat /tmp/err32.json; echo
 [ "$CODE" = "503" ] && ok "503 bei downem TTS-Worker" || bad "erwartet 503, bekam $CODE"
 RESP=$(curl -sf "$BASE/status")
@@ -634,7 +634,7 @@ MOCK_TTS_PORT="$TTS_PORT" MOCK_TTS_DELAY=0.1 \
 TTS_PID=$!
 sleep 0.5
 CODE=$(curl -s -o /tmp/tts33.wav -w "%{http_code}" "$BASE/v1/audio/speech" \
-  -H "Content-Type: application/json" -d '{"input":"Wieder da","voice":"claribel","response_format":"wav"}')
+  -H "Content-Type: application/json" -d '{"input":"Wieder da","voice":"alloy","response_format":"wav"}')
 [ "$CODE" = "200" ] && [ -s /tmp/tts33.wav ] \
  && ok "TTS nach Neustart wieder verfügbar" || bad "TTS-Recovery (Code $CODE)"

@@ -727,8 +727,8 @@ import json,sys
 d=json.load(sys.stdin)
 ids={m["id"] for m in d["data"]}
 assert "whisper-1" in ids, ids
-assert "xtts-v2" in ids, ids
-' && ok "Audio-Modelle: whisper-1 + xtts-v2" || bad "Audio-Modelle"
+assert "qwen3-tts" in ids, ids
+' && ok "Audio-Modelle: whisper-1 + qwen3-tts" || bad "Audio-Modelle"

 # --- 41. /v1/audio/voices ------------------------------------------------------------------------------------------
 echo "== Test 41: GET /v1/audio/voices"
@@ -738,8 +738,8 @@ echo "$RESP" | python3 -c '
 import json,sys
 d=json.load(sys.stdin)
 ids={v["id"] for v in d["data"]}
-assert "claribel" in ids, ids
-' && ok "Audio-Voices: claribel" || bad "Audio-Voices"
+assert "alloy" in ids, ids
+' && ok "Audio-Voices: alloy" || bad "Audio-Voices"

 # --- 42. STT + Qwen parallel ----------------------------------------------------------------------------------------
 echo "== Test 42: STT + Qwen parallel"
@@ -770,7 +770,7 @@ sleep 0.2
 # TTS-Request
 CODE=$(curl -s -o /tmp/tts43.mp3 -w "%{http_code}" \
  "$BASE/v1/audio/speech" -H "Content-Type: application/json" \
-  -d '{"input":"Hallo","voice":"claribel"}')
+  -d '{"input":"Hallo","voice":"alloy"}')
 wait $STT_PID43
 [ "$CODE" = "200" ] && [ -s /tmp/tts43.mp3 ] \
  && ok "STT + TTS parallel (beide 200)" || bad "STT + TTS parallel (TTS Code $CODE)"
@@ -8,7 +8,7 @@ flowchart LR
    P --> Q[genau ein llama.cpp-Profil<br/>Qwen Fast / Medium / Large / Ultra / Uncensored]
    R --> I[FLUX.2 Klein 9B FP8 Beta<br/>RTX 5080 Transformer]
    I --> E[Qwen3-8B NF4 Textencoder<br/>RTX 3060 während Bildauftrag]
-    R --> T[Qwen3-TTS RTX 3060<br/>Piper CPU-Fallback]
+    R --> T[Qwen3-TTS RTX 3060<br/>Normalisierungs- und Streaming-Gateway]
    R --> STT[Whisper.cpp ggml-small<br/>CPU, lokale Spracherkennung]

    H --> U[MUA / Unraid MCP]
@@ -50,7 +50,6 @@ Kontextgröße:
 | Medium | 160.000 Token |
 | Large | 192.000 Token |
 | Ultra | 262.144 Token |
-| Beta 1 | 112.000 Token |
 | Uncensored | 80.000 Token |

 ## Exklusiver Bildmodus
@@ -65,7 +64,7 @@ gleichzeitig. Der Wechsel ist transaktional:
 4. Nach dem Prompt-Encoding werden die Embeddings zur RTX 5080 übertragen.
 5. Vor dem VAE-Decoding werden Textencoder und Transformer freigegeben.
 6. Der Worker wird gestoppt; anschließend starten Qwen3-TTS und das vorherige
-   Textprofil wieder. Piper bleibt als CPU-Fallback verfügbar.
+   Textprofil wieder. Während der exklusiven Bildphase steht kein TTS bereit.

 Der Bild-Worker ist lazy und besitzt `restart: "no"`; im normalen Textbetrieb
 belegt er daher keinen VRAM. Container werden über eindeutige Docker-Labels
@@ -1,8 +1,8 @@
 # Container-Inventar auf Athena

-Stand: 9. September 2026
+Stand: 10. September 2026

-Athena besteht derzeit aus 25 Docker-Containern. Nicht jeder Container enthält
+Athena besteht nach der Bereinigung aus 23 Docker-Containern. Nicht jeder Container enthält
 ein KI-Modell: Router, Oberflächen, Netzwerk, Steuerung und Sicherung sind
 gewöhnliche Dienste. Die rechenintensiven GPU-Worker werden absichtlich nur bei
 Bedarf gestartet. Ein Container im Zustand `Created` oder `Exited (0)` ist daher
@@ -13,7 +13,6 @@ nicht automatisch ein ungenutzter Rest.
 | `mike-ai-backup` | kein Modell; Offen Docker Volume Backup | Sichert `/data`, `/etc/mike-ai`, den Stack und die persistenten Docker-Volumes im Fünf-Stunden-Takt. |
 | `mike-ai-applio-studio` | Applio/RVC; Stimmenmodelle werden nutzerseitig ergänzt | Vollständige RVC-Oberfläche für Inferenz, Modellverwaltung und Training auf der RTX 5080. Für eine Konvertierung ist ein importiertes oder trainiertes `.pth`-Modell nötig; eine Referenzaufnahme allein reicht nicht. |
 | `mike-ai-image-worker` | FLUX.2 Klein 9B FP8, Qwen3-8B NF4 Textencoder und VAE | Erzeugt und bearbeitet Bilder transaktional; nutzt während eines Auftrags RTX 5080 und RTX 3060. |
-| `mike-ai-llama-beta1` | Qwen3.8-27B GSQ-RCO `IQ3_S-MTP`, Qwen-MMProj BF16 | Experimentelles Beta-Profil mit 112.000 Token Kontext; wird nur auf Anforderung geladen. |
 | `mike-ai-llama-dashboard` | kein Modell | Zeigt Telemetrie, Profile, GPU-Nutzung und Betriebsarten an und bietet die Modusumschaltung. |
 | `mike-ai-llama-fast` | Qwen3.8-27B `IQ4-MIX`, Qwen-MMProj BF16 | Schnelles Q4-Text-/Vision-Profil mit 76.800 Token Kontext. |
 | `mike-ai-llama-large` | Qwen3.8-27B `IQ4_XS-pure`, Qwen-MMProj BF16 | Q4-Text-/Vision-Profil mit 192.000 Token Kontext und Verteilung auf beide GPUs. |
@@ -23,13 +22,12 @@ nicht automatisch ein ungenutzter Rest.
 | `mike-ai-mcp-athena-operator` | kein Modell | Stellt Hermes begrenzte Werkzeuge zum Prüfen, Ändern, Testen, Sichern und Versionieren von Athena bereit. |
 | `mike-ai-music-acestep-test` | ACE-Step 1.5 XL-SFT und `acestep-5Hz-lm-1.7B` | Generiert Musik im exklusiven Musikmodus auf der RTX 5080. |
 | `mike-ai-music-ui` | kein Modell; `fspecii/ace-step-ui` | Community-Oberfläche für ACE-Step; bleibt als leichte UI verfügbar, während der GPU-Worker bedarfsgesteuert läuft. |
-| `mike-ai-piper` | `de_DE-thorsten-high` | CPU-basierte deutsche TTS-Rückfallebene, die auch während GPU-Umschaltungen verfügbar bleibt. |
 | `mike-ai-portainer` | kein Modell; Portainer CE | Optionale Docker-Verwaltungsoberfläche. |
 | `mike-ai-profile-controller` | kein Modell | Startet und stoppt ausschließlich freigegebene Modellprofile und Spezialworker in einer sicheren Reihenfolge. |
 | `mike-ai-qwen3-tts` | `Qwen/Qwen3-TTS-12Hz-1.7B-Base`, Stimme Serena | Hochwertige deutsche Sprachausgabe auf der RTX 3060 im LLM-Betrieb. |
 | `mike-ai-router` | kein eigenes Modell | Einzige OpenAI-kompatible Modelladresse; koordiniert Profile, Bildaufträge, Sprache und Betriebsarten. |
 | `mike-ai-stem-separator` | BS-RoFormer Viperx 1297, `htdemucs_ft`, `htdemucs_6s`, `MossFormer2_SE_48K` | Trennt Gesang, Instrumente oder Sprache/Hintergrundgeräusche im exklusiven Separationsmodus. |
-| `mike-ai-tts-gateway` | kein eigenes Modell | Normalisiert Text, leitet TTS an Qwen3-TTS weiter und fällt bei Bedarf auf Piper zurück. |
+| `mike-ai-tts-gateway` | kein eigenes Modell | Normalisiert Text, konvertiert Ausgabeformate und stellt Qwen3-TTS sowie natives PCM-Streaming über eine stabile interne API bereit. |
 | `mike-ai-voice-studio` | `k2-fsa/OmniVoice` 0.2.1 mit Whisper-ASR | Erzeugt Text-to-Speech mit einer Referenzstimme; kein Audio-to-Audio-Voice-Changer. |
 | `mike-ai-whisper` | Whisper.cpp `ggml-small` | Lokale deutsche Spracherkennung auf der CPU über `/v1/audio/transcriptions`. |
 | `mike-ai-wireguard-gateway` | kein Modell | Veröffentlicht Dashboard und Fachoberflächen ausschließlich über den privaten WireGuard-Pfad. |
@@ -24,7 +24,7 @@ bereit:
 - RTX 5080: 9B-FP8-Diffusionstransformer und VAE-Decoding
 - RTX 3060: Qwen3-8B-Textencoder in NF4
 - Qwen3-TTS und aktives llama.cpp-Profil werden für den Bildauftrag pausiert
- Piper bleibt währenddessen als CPU-TTS verfügbar
+- TTS ist währenddessen vorübergehend nicht verfügbar
 - nach Abschluss werden TTS und das vorherige Textprofil wiederhergestellt

 Ein vollständiger Aufruf über Athenas OpenAI-kompatiblen Router wurde mit
@@ -61,7 +61,7 @@ Build 10718, Commit `41ef91f7c8046087cdfbb276b79bff311ecf1c6d`. Dessen lokales
 Fallback-Image wurde am 8. September 2026 beim gezielten Aufräumen entfernt;
 ein Rückfall erfordert daher einen Neubau dieses Commits.

-Build 10781 wurde nach dem Bau produktiv verifiziert. Alle sechs
+Build 10781 wurde nach dem Bau produktiv verifiziert. Alle fünf
 Profildefinitionen verwenden dasselbe neue Image. Am 8. September lief Ultra
 mit 262.144 Tokens Kontext gesund; MTP und eine lokale Textprobe wurden
 erfolgreich geprüft.
@@ -44,7 +44,7 @@ Der Profile Controller stoppt vor dem Start des Bild-Workers alle
 llama.cpp-Profile und den mit `com.mike-ai.tts-worker=qwen3` markierten
 Qwen3-TTS-Container. Dadurch bleibt genügend VRAM für beide Bildkomponenten.
 Nach dem Bildauftrag startet er Qwen3-TTS und das zuvor aktive Textprofil
-wieder. Der TTS-Gateway kann während des Wechsels auf Piper ausweichen.
+wieder. Während des exklusiven GPU-Wechsels ist TTS vorübergehend nicht verfügbar.

 ## Aktuelle Grenzen

@@ -1,7 +1,12 @@
 # Qwen Beta 1 – GSQ-RCO

-`qwen-beta-1` ist ein zusätzliches, nicht standardmäßig aktives Router-Profil.
-Die bestehenden Profile und das Standardprofil `qwen-medium` bleiben unverändert.
+> Historischer Testbericht. Das Beta-1-Profil wurde am 10. September 2026
+> vollständig aus dem produktiven Router entfernt, weil die kleinere
+> Quantisierung gegenüber den Q4-Profilen keinen belastbaren Vorteil brachte.
+
+`qwen-beta-1` war ein zusätzliches, nicht standardmäßig aktives Router-Profil.
+Der Bericht bleibt erhalten, damit diese Quantisierung nicht versehentlich
+erneut getestet wird.

 ## Laufzeitkonfiguration

@@ -7,7 +7,6 @@

 - `/etc/mike-ai` mit lokaler Konfiguration,
 - Router-Zustand und erzeugte Bilder,
- Piper-Daten,
 - der kanonische Stack als zusätzlicher Snapshot.

 Nicht in das Archiv gehören die großen Modellgewichte unter `/data/models`.
@@ -8,7 +8,6 @@ Standardprofil: **medium** · globales Ausgabelimit: **8192 Token**
 |---|---|---:|---:|---|---|---|---:|
 | fast | `qwen-fast` | 76,800 | 1 | Qwen3.8-27B IQ4 Mix | 5080 only | ja | 2 |
 | medium | `qwen-medium` | 160,000 | 1 | Qwen3.8-27B IQ4 XS Pure | 85:15 | ja | 3 |
-| beta1 | `qwen-beta-1` | 112,000 | 1 | Qwen3.8-27B GSQ-RCO IQ3_S MTP | 5080 model / 3060 vision | ja | 3 |
 | large | `qwen-large` | 192,000 | 1 | Qwen3.8-27B IQ4 XS Pure | 86:14 | ja | 3 |
 | ultra | `qwen-ultra` | 262,144 | 1 | Qwen3.8-27B IQ4 XS Pure | 80:20 | nein | 2 |
 | uncensored | `qwen-uncensored` | 80,000 | 1 | Qwen3.8-27B Abliterated Q4_K_M | 90:10 | ja | 2 |
@@ -17,7 +16,6 @@ Standardprofil: **medium** · globales Ausgabelimit: **8192 Token**

 - **fast**: Schnelles Profil für kurze Chats und zügige Werkzeugaufgaben.
 - **medium**: Ausgewogenes Standardprofil für Alltag und lange agentische Aufgaben.
- **beta1**: Beta 1: qualitaetsoptimiertes GSQ-RCO-IQ3_S-Testprofil; 112K Startkontext bis zur erneuten Grenzmessung.
 - **large**: Großes Profil für umfangreiche Dokumente und lange technische Arbeiten.
 - **ultra**: Maximaler Textkontext; bewusst ohne Vision-Projektor.
 - **uncensored**: Weniger restriktives Spezialprofil; Werkzeugrechte bleiben unverändert.
@@ -27,7 +27,7 @@ Statuswerte:
 | 04.09.2026 | ISTA-DASLab Qwen3.8-27B GSQ-RCO `IQ3_XXS-MTP` | bis 196.608 | sehr platzsparend und bis 196.608 technisch lauffähig; später durch IQ3_S ersetzt | **ersetzt** | [GSQ_RCO_BETA1_20260904.md](GSQ_RCO_BETA1_20260904.md) |
 | 07.09.2026 | `Jackrong/Qwopus3.8-27B-Flash-GGUF` / `Qwopus3.8-27B-Flash-MTP-IQ4_XS.gguf` | 160K | Recall 3/3; Decode 87,2 statt 105,3 Token/s, Lang-Decode 51,0 statt 56,7 Token/s; kein Gesamtvorteil | **verworfen** | Athena: `/data/benchmarks/qwen38-ab-20260907/` |
 | 07.09.2026 | `bartowski/Qwen3.8-27B-GGUF` / `Qwen3.8-27B-IQ4_XS.gguf` | 160K | Recall 3/3; Decode 90,4 statt 105,3 Token/s, Lang-Decode 51,9 statt 56,7 Token/s; kein Gesamtvorteil | **verworfen** | Athena: `/data/benchmarks/qwen38-ab-20260907/` |
-| 08.09.2026 | `ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF` / `IQ3_S-MTP` | 76,8K–262K | Qualität im lokalen Test praktisch gleich, trotz optimierter GPU-Splits überwiegend langsamer als Q4 | **Beta** mit 112K; kein Ersatz für Q4 | [GSQ_RCO_BETA1_20260904.md](GSQ_RCO_BETA1_20260904.md) |
+| 08.09.2026 | `ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF` / `IQ3_S-MTP` | 76,8K–262K | Qualität im lokalen Test praktisch gleich, trotz optimierter GPU-Splits überwiegend langsamer als Q4 | **entfernt**; kein Ersatz für Q4 | [GSQ_RCO_BETA1_20260904.md](GSQ_RCO_BETA1_20260904.md) |
 | 08.09.2026 | `Tiel-Coder-35B-A3B-UD-IQ4_XS.gguf` | 160K vorgesehen | Testcontainer und Gewichte vorhanden gewesen, aber kein versionierter, belastbarer Abnahmebericht | **unvollständig**; nicht als getesteter Sieger behandeln | kein Ergebnisartefakt vorhanden |

 ## Externe CPU-Helfermodelle
@@ -54,7 +54,7 @@ Titelgenerierung und Kontextkompression in Hermes.
 | Datum | Modell | Ergebnis | Status / Entscheidung |
 |---|---|---|---|
 | 05.09.2026 | Coqui XTTS v2 | deutsche Satzzeichen, Enden und Streaming-Chunks erzeugten Halluzinationen und unnatürliche Prosodie | **verworfen und entfernt** |
-| 05.09.2026 | `Qwen/Qwen3-TTS-12Hz-1.7B-Base` | deutlich natürlichere deutsche Ausgabe ohne die XTTS-Endhalluzinationen | **produktiv** mit Piper-Fallback |
+| 05.09.2026 | `Qwen/Qwen3-TTS-12Hz-1.7B-Base` | deutlich natürlichere deutsche Ausgabe ohne die XTTS-Endhalluzinationen | **produktiv** als einziges TTS-Backend |
 | 06.–08.09.2026 | Whisper.cpp `large-v3-turbo` | lokaler TTS→STT-Rundlauf und OpenClaw-Transkription erfolgreich; später zugunsten des kleineren Laufzeitmodells entfernt | **ersetzt** |
 | seit 03.09.2026 | Whisper.cpp `ggml-small` | tatsächlich im Compose-Stack und im laufenden Container verwendetes CPU-STT-Modell | **produktiv** |
 | 09.09.2026 | `RMSnow/Vevo2`, Amphion `26f6883110181f1dbfe95c70a7c7dbaf4de5f42a` | Technik und Geschwindigkeit funktionierten, reale deutsche Sprachwandlung mit kurzer und langer Referenz war jedoch unverständlich, halluzinierend oder musikalisch | **qualitativ verworfen und entfernt**; Ergebnis bleibt hier dokumentiert, Images, Daten und altes Projekt wurden am 09.09. bereinigt |
@@ -41,7 +41,6 @@ source "$CONFIG"
 required=(AI_HOSTNAME ADMIN_USER MODEL_DIR FAST_MODEL_FILE
  FAST_MODEL_URL FAST_MODEL_SHA256 MEDIUM_MODEL_FILE MEDIUM_MODEL_URL
  MEDIUM_MODEL_SHA256 LARGE_MODEL_FILE LARGE_MODEL_URL LARGE_MODEL_SHA256
-  BETA1_MODEL_FILE BETA1_MODEL_URL BETA1_MODEL_SHA256
  ULTRA_MODEL_FILE ULTRA_MODEL_URL ULTRA_MODEL_SHA256
  UNCENSORED_MODEL_FILE UNCENSORED_MODEL_URL UNCENSORED_MODEL_SHA256
  UNCENSORED_PROJECTOR_FILE UNCENSORED_PROJECTOR_URL
@@ -314,8 +313,6 @@ MODEL_DIR=$MODEL_DIR
 WIREGUARD_CONFIG_FILE=${WIREGUARD_CONFIG_FILE:-/etc/mike-ai/wireguard/fritz-athena.conf}
 ROUTER_API_KEY=$(<$SECRETS_DIR/router-api-key)
 CONTROLLER_TOKEN=$(<$SECRETS_DIR/controller-token)
-PIPER_TTS_VERSION=${PIPER_TTS_VERSION:-1.6.0}
-PIPER_VOICE=${PIPER_VOICE:-de_DE-thorsten-high}
 QWEN3_TTS_IMAGE=${QWEN3_TTS_IMAGE:-ghcr.io/malaiwah/qwen3-tts-server:latest@sha256:b363a01d08b1bbecbfc3ca6f585368fae2cfdc591f9ecca6643738369f9a9d98}
 QWEN3_TTS_CACHE_DIR=$QWEN3_TTS_CACHE_DIR
 QWEN3_TTS_VOICES_DIR=$QWEN3_TTS_VOICES_DIR
@@ -323,7 +320,6 @@ QWEN3_TTS_GPU_DEVICE=${QWEN3_TTS_GPU_DEVICE:-GPU-4834d9d7-5b61-3004-1fb3-4ae49d4
 AI_DNS=${WG_DNS:-1.1.1.1}
 FAST_MODEL_FILE=$FAST_MODEL_FILE
 MEDIUM_MODEL_FILE=$MEDIUM_MODEL_FILE
-BETA1_MODEL_FILE=$BETA1_MODEL_FILE
 LARGE_MODEL_FILE=$LARGE_MODEL_FILE
 ULTRA_MODEL_FILE=$ULTRA_MODEL_FILE
 UNCENSORED_MODEL_FILE=$UNCENSORED_MODEL_FILE
@@ -337,10 +333,6 @@ MEDIUM_CONTEXT=${MEDIUM_CONTEXT:-160000}
 MEDIUM_BATCH_SIZE=${MEDIUM_BATCH_SIZE:-2048}
 MEDIUM_UBATCH_SIZE=${MEDIUM_UBATCH_SIZE:-128}
 MEDIUM_PARALLEL_SLOTS=${MEDIUM_PARALLEL_SLOTS:-1}
-BETA1_CONTEXT=${BETA1_CONTEXT:-112000}
-BETA1_BATCH_SIZE=${BETA1_BATCH_SIZE:-2048}
-BETA1_UBATCH_SIZE=${BETA1_UBATCH_SIZE:-128}
-BETA1_PARALLEL_SLOTS=${BETA1_PARALLEL_SLOTS:-1}
 LARGE_CONTEXT=${LARGE_CONTEXT:-192000}
 LARGE_BATCH_SIZE=${LARGE_BATCH_SIZE:-2048}
 LARGE_UBATCH_SIZE=${LARGE_UBATCH_SIZE:-128}
@@ -356,7 +348,6 @@ UNCENSORED_PARALLEL_SLOTS=${UNCENSORED_PARALLEL_SLOTS:-1}
 FAST_GPU_DEVICES=${TEXT_GPU_DEVICES:-0}${SECONDARY_GPU_DEVICES:+,$SECONDARY_GPU_DEVICES}
 MEDIUM_GPU_DEVICES=${TEXT_GPU_DEVICES:-0}${SECONDARY_GPU_DEVICES:+,$SECONDARY_GPU_DEVICES}
 MEDIUM_TENSOR_SPLIT=${MEDIUM_TENSOR_SPLIT:-90,10}
-BETA1_GPU_DEVICES=${TEXT_GPU_DEVICES:-0}${SECONDARY_GPU_DEVICES:+,$SECONDARY_GPU_DEVICES}
 LARGE_GPU_DEVICES=${TEXT_GPU_DEVICES:-0}${SECONDARY_GPU_DEVICES:+,$SECONDARY_GPU_DEVICES}
 LARGE_TENSOR_SPLIT=${LARGE_TENSOR_SPLIT:-86,14}
 ULTRA_GPU_DEVICES=${TEXT_GPU_DEVICES:-0}${SECONDARY_GPU_DEVICES:+,$SECONDARY_GPU_DEVICES}
@@ -405,7 +396,6 @@ download_models() {
  done <<EOF
 $FAST_MODEL_FILE|$FAST_MODEL_URL|$FAST_MODEL_SHA256
 $MEDIUM_MODEL_FILE|$MEDIUM_MODEL_URL|$MEDIUM_MODEL_SHA256
-$BETA1_MODEL_FILE|$BETA1_MODEL_URL|$BETA1_MODEL_SHA256
 $LARGE_MODEL_FILE|$LARGE_MODEL_URL|$LARGE_MODEL_SHA256
 $ULTRA_MODEL_FILE|$ULTRA_MODEL_URL|$ULTRA_MODEL_SHA256
 $UNCENSORED_MODEL_FILE|$UNCENSORED_MODEL_URL|$UNCENSORED_MODEL_SHA256
@@ -514,11 +504,11 @@ build_and_start() {
  # Portable MCPs and Hermes live on Unraid and are restored through Appdata.
  "$STACK_DIR/platform/mcp/install-tools.sh"
  docker compose --env-file "$SECRETS_DIR/stack.env" --profile inference create \
-    llama-fast llama-medium llama-beta1 llama-large llama-ultra llama-uncensored
+    llama-fast llama-medium llama-large llama-ultra llama-uncensored
  docker compose --env-file "$SECRETS_DIR/stack.env" --profile image create image-worker
  docker volume inspect portainer_data >/dev/null 2>&1 || docker volume create portainer_data >/dev/null
  docker compose --env-file "$SECRETS_DIR/stack.env" up -d --build \
-    wireguard-gateway qwen3-tts piper tts-gateway profile-controller router llama-dashboard portainer backup
+    wireguard-gateway qwen3-tts tts-gateway profile-controller router llama-dashboard portainer backup
  if [[ ${WIREGUARD_MODE:-container} == container ]]; then
    systemctl restart mike-ai-container-vpn-guard.service
  fi
@@ -7,7 +7,7 @@ Athena speech stack:
 2. Athena Whisper transcribes it,
 3. OpenClaw's normal agent-consult path answers with its configured model and
   tools,
-4. Athena XTTS/Piper returns PCM audio to the Talk client.
+4. Athena Qwen3-TTS returns PCM audio to the Talk client.

 Long replies are synthesized incrementally. The first short phrase starts
 playing as soon as it is ready while the next phrase is generated in parallel.
@@ -324,7 +324,7 @@ export default definePluginEntry({
            label: "Athena Local Talk",
            aliases: ["athena", "local-athena"],
            defaultModel: "athena-local",
-            voices: ["alloy", "claribel"],
+            voices: ["alloy"],
            autoSelectOrder: 1,
            capabilities: {
                transports: ["gateway-relay"],
@@ -307,7 +307,7 @@ class AthenaTalkBridge {
    const response = await fetch(`${this.cfg.baseUrl}/audio/speech`, {
      method: "POST",
      headers: { "Content-Type": "application/json", ...this.authHeaders() },
-      // Let Athena select its currently active local backend (XTTS or Piper).
+      // Athena normalizes the request and serves it through Qwen3-TTS.
      body: JSON.stringify({ voice: this.cfg.voice, input: text, response_format: "wav" }),
      signal,
    });
@@ -330,7 +330,7 @@ export default definePluginEntry({
      label: "Athena Local Talk",
      aliases: ["athena", "local-athena"],
      defaultModel: "athena-local",
-      voices: ["alloy", "claribel"],
+      voices: ["alloy"],
      autoSelectOrder: 1,
      capabilities: {
        transports: ["gateway-relay"],
@@ -52,7 +52,7 @@ case "$command" in
      [[ $# -eq 0 ]] || { echo "core akzeptiert keine weiteren Services" >&2; exit 2; }
      run "$ROOT_DIR/platform/mcp/install-tools.sh"
      run "${compose[@]}" up -d --build \
-        wireguard-gateway piper xtts tts-gateway profile-controller router llama-dashboard portainer backup
+        wireguard-gateway qwen3-tts tts-gateway profile-controller router llama-dashboard portainer backup
    else
      run "${compose[@]}" up -d --build --no-deps "$@"
    fi
@@ -1,24 +0,0 @@
-FROM python:3.12-slim-bookworm
-
-ARG PIPER_TTS_VERSION=1.6.0
-
-RUN apt-get update \
-    && apt-get install -y --no-install-recommends ca-certificates curl ffmpeg gosu \
-    && python -m pip install --no-cache-dir "piper-tts==${PIPER_TTS_VERSION}" \
-    && useradd --system --uid 10003 --home-dir /nonexistent --shell /usr/sbin/nologin piper \
-    && rm -rf /var/lib/apt/lists/*
-
-WORKDIR /app
-COPY piper_worker.py /app/piper_worker.py
-COPY entrypoint.sh /usr/local/bin/mike-ai-piper-entrypoint
-RUN chmod 0755 /usr/local/bin/mike-ai-piper-entrypoint
-
-ENV PIPER_DATA_DIR=/data \
-    PIPER_VOICE=de_DE-thorsten-high \
-    PIPER_VOICE_ALIAS=alloy \
-    PIPER_HOST=0.0.0.0 \
-    PIPER_PORT=8085
-
-VOLUME ["/data"]
-EXPOSE 8085
-ENTRYPOINT ["/usr/local/bin/mike-ai-piper-entrypoint"]
@@ -1,15 +0,0 @@
-#!/bin/sh
-set -eu
-
-data_dir=${PIPER_DATA_DIR:-/data}
-voice=${PIPER_VOICE:-de_DE-thorsten-high}
-
-mkdir -p "$data_dir"
-chown 10003:10003 "$data_dir"
-
-if [ ! -s "$data_dir/$voice.onnx" ] || [ ! -s "$data_dir/$voice.onnx.json" ]; then
-  echo "Downloading Piper voice: $voice"
-  gosu piper python -m piper.download_voices --data-dir "$data_dir" "$voice"
-fi
-
-exec gosu piper python /app/piper_worker.py
@@ -1,153 +0,0 @@
-#!/usr/bin/env python3
-"""Small, private Piper worker for the Mike AI profile router.
-
-The public OpenAI-compatible endpoint remains in the router.  This worker only
-accepts the narrow internal /status and /tts protocol and never logs input text.
-"""
-
-from __future__ import annotations
-
-import io
-import json
-import os
-import subprocess
-import threading
-import wave
-from http import HTTPStatus
-from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
-from pathlib import Path
-
-from piper import PiperVoice, SynthesisConfig
-
-
-DATA_DIR = Path(os.getenv("PIPER_DATA_DIR", "/data"))
-VOICE_NAME = os.getenv("PIPER_VOICE", "de_DE-thorsten-high")
-VOICE_ALIAS = os.getenv("PIPER_VOICE_ALIAS", "alloy")
-HOST = os.getenv("PIPER_HOST", "0.0.0.0")
-PORT = int(os.getenv("PIPER_PORT", "8085"))
-MAX_TEXT_CHARS = int(os.getenv("PIPER_MAX_TEXT_CHARS", "8000"))
-MAX_REQUEST_BYTES = int(os.getenv("PIPER_MAX_REQUEST_BYTES", "65536"))
-
-VOICE_PATH = DATA_DIR / f"{VOICE_NAME}.onnx"
-VOICE = PiperVoice.load(str(VOICE_PATH))
-SYNTHESIS_LOCK = threading.Lock()
-
-
-def synthesize_wav(text: str, speed: float) -> bytes:
-    """Synthesize a complete WAV in memory without retaining the text."""
-    output = io.BytesIO()
-    config = SynthesisConfig(length_scale=1.0 / speed)
-    with SYNTHESIS_LOCK, wave.open(output, "wb") as wav_file:
-        VOICE.synthesize_wav(text, wav_file, syn_config=config)
-    return output.getvalue()
-
-
-def wav_to_mp3(wav_bytes: bytes) -> bytes:
-    """Convert Piper's WAV to the MP3 format Open WebUI requests by default."""
-    result = subprocess.run(
-        [
-            "ffmpeg", "-hide_banner", "-loglevel", "error",
-            "-f", "wav", "-i", "pipe:0",
-            "-codec:a", "libmp3lame", "-b:a", "96k",
-            "-f", "mp3", "pipe:1",
-        ],
-        input=wav_bytes,
-        stdout=subprocess.PIPE,
-        stderr=subprocess.PIPE,
-        check=False,
-        timeout=120,
-    )
-    if result.returncode != 0:
-        raise RuntimeError("ffmpeg conversion failed")
-    return result.stdout
-
-
-class Handler(BaseHTTPRequestHandler):
-    protocol_version = "HTTP/1.1"
-
-    def log_message(self, fmt: str, *args: object) -> None:
-        # Deliberately omit URLs and request bodies from the log.
-        print(f"piper-worker: {self.command} -> {args[1] if len(args) > 1 else '-'}")
-
-    def send_bytes(self, status: int, body: bytes, content_type: str) -> None:
-        self.send_response(status)
-        self.send_header("Content-Type", content_type)
-        self.send_header("Content-Length", str(len(body)))
-        self.send_header("Cache-Control", "no-store")
-        self.end_headers()
-        self.wfile.write(body)
-
-    def send_json(self, status: int, payload: dict) -> None:
-        self.send_bytes(
-            status,
-            json.dumps(payload, separators=(",", ":")).encode(),
-            "application/json",
-        )
-
-    def do_GET(self) -> None:  # noqa: N802
-        if self.path != "/status":
-            self.send_json(HTTPStatus.NOT_FOUND, {"error": "not found"})
-            return
-        self.send_json(
-            HTTPStatus.OK,
-            {
-                "ready": True,
-                "engine": "piper",
-                "model": VOICE_NAME,
-                "voices": [VOICE_ALIAS],
-            },
-        )
-
-    def do_POST(self) -> None:  # noqa: N802
-        if self.path != "/tts":
-            self.send_json(HTTPStatus.NOT_FOUND, {"error": "not found"})
-            return
-
-        try:
-            content_length = int(self.headers.get("Content-Length", "0"))
-        except ValueError:
-            content_length = 0
-        if content_length <= 0 or content_length > MAX_REQUEST_BYTES:
-            self.send_json(HTTPStatus.REQUEST_ENTITY_TOO_LARGE, {"error": "invalid request size"})
-            return
-
-        try:
-            request = json.loads(self.rfile.read(content_length))
-            text = request.get("text", "")
-            voice = request.get("voice", VOICE_ALIAS)
-            output_format = request.get("format", "mp3")
-            speed = float(request.get("speed", 1.0))
-        except (json.JSONDecodeError, TypeError, ValueError):
-            self.send_json(HTTPStatus.BAD_REQUEST, {"error": "invalid JSON request"})
-            return
-
-        if not isinstance(text, str) or not text.strip() or len(text) > MAX_TEXT_CHARS:
-            self.send_json(HTTPStatus.BAD_REQUEST, {"error": "invalid text"})
-            return
-        if voice != VOICE_ALIAS:
-            self.send_json(HTTPStatus.BAD_REQUEST, {"error": "unknown voice"})
-            return
-        if output_format not in {"wav", "mp3"}:
-            self.send_json(HTTPStatus.BAD_REQUEST, {"error": "unsupported format"})
-            return
-        if not 0.5 <= speed <= 2.0:
-            self.send_json(HTTPStatus.BAD_REQUEST, {"error": "invalid speed"})
-            return
-
-        try:
-            audio = synthesize_wav(text.strip(), speed)
-            if output_format == "mp3":
-                audio = wav_to_mp3(audio)
-                content_type = "audio/mpeg"
-            else:
-                content_type = "audio/wav"
-        except (OSError, RuntimeError, subprocess.SubprocessError):
-            self.send_json(HTTPStatus.INTERNAL_SERVER_ERROR, {"error": "synthesis failed"})
-            return
-
-        self.send_bytes(HTTPStatus.OK, audio, content_type)
-
-
-if __name__ == "__main__":
-    print(f"Piper worker ready: {VOICE_NAME} as {VOICE_ALIAS} on {HOST}:{PORT}")
-    ThreadingHTTPServer((HOST, PORT), Handler).serve_forever()
@@ -230,7 +230,7 @@ def set_image_worker(running: bool, kind: str = IMAGE_WORKER) -> dict:
            for profile_item in containers().values():
                stop_container(profile_item)
            # The 9B beta text encoder temporarily borrows the RTX 3060 from
-            # Qwen3-TTS. The gateway retains Piper as a fallback meanwhile.
+            # Qwen3-TTS. TTS is unavailable during this exclusive GPU phase.
            stop_container(tts_container(), timeout=30)
            stop_music_if_configured()
            stop_separator_if_configured()
@@ -244,25 +244,20 @@ class LanguageSegmentationTests(unittest.TestCase):
        self.assertTrue(all(len(part.split()) > 4 for _, part in segments))


-class FallbackTests(unittest.TestCase):
+class BackendFailureTests(unittest.TestCase):
    def setUp(self):
-        self.original_xtts = gateway.synthesize_xtts
-        self.original_piper = gateway.synthesize_piper
+        self.original_qwen = gateway.synthesize_qwen

    def tearDown(self):
-        gateway.synthesize_xtts = self.original_xtts
-        gateway.synthesize_piper = self.original_piper
+        gateway.synthesize_qwen = self.original_qwen

-    def test_piper_is_used_when_xtts_fails(self):
+    def test_qwen_failure_is_reported_without_fallback(self):
        def fail(*_args):
-            raise RuntimeError("synthetic XTTS failure")
+            raise RuntimeError("synthetic Qwen failure")

-        gateway.synthesize_xtts = fail
-        gateway.synthesize_piper = lambda *_args: (b"piper", "audio/wav")
-        self.assertEqual(
-            gateway.synthesize("synthetic test", "wav", 1.0),
-            (b"piper", "audio/wav"),
-        )
+        gateway.synthesize_qwen = fail
+        with self.assertRaisesRegex(RuntimeError, "synthetic Qwen failure"):
+            gateway.synthesize("synthetic test", "wav", 1.0)


 class AudioJoinTests(unittest.TestCase):
@@ -1,5 +1,5 @@
 #!/usr/bin/env python3
-"""Private Qwen3-TTS-first gateway with a Piper fallback.
+"""Private Qwen3-TTS gateway.

 The gateway implements the narrow /status and /tts protocol already consumed
 by the profile router. Request text is never logged or persisted.
@@ -33,7 +33,6 @@ QWEN_TTS_VOICE = os.getenv("QWEN_TTS_VOICE", "serena")
 QWEN_TTS_LANGUAGE = os.getenv("QWEN_TTS_LANGUAGE", "German")
 QWEN_TTS_TIMEOUT = float(os.getenv("QWEN_TTS_TIMEOUT", "120"))
 XTTS_URL = os.getenv("XTTS_URL", "http://xtts:80").rstrip("/")
-PIPER_URL = os.getenv("PIPER_URL", "http://piper:8085").rstrip("/")
 VOICE_ALIAS = os.getenv("TTS_VOICE_ALIAS", "alloy")
 XTTS_SPEAKER = os.getenv("XTTS_SPEAKER", "Annmarie Nele")
 DEFAULT_LANGUAGE = os.getenv("TTS_DEFAULT_LANGUAGE", "de")
@@ -43,7 +42,6 @@ MAX_TEXT_CHARS = int(os.getenv("TTS_MAX_TEXT_CHARS", "8000"))
 MAX_REQUEST_BYTES = int(os.getenv("TTS_MAX_REQUEST_BYTES", "65536"))
 MAX_AUDIO_BYTES = int(os.getenv("TTS_MAX_AUDIO_BYTES", str(64 * 1024 * 1024)))
 XTTS_TIMEOUT = float(os.getenv("XTTS_TIMEOUT", "120"))
-PIPER_TIMEOUT = float(os.getenv("PIPER_TIMEOUT", "120"))
 QUEUE_TIMEOUT = float(os.getenv("XTTS_QUEUE_TIMEOUT", "15"))
 # XTTS loses natural prosody when a sentence is synthesized as many tiny
 # requests: every request starts a fresh utterance. Keep complete sentences
@@ -63,8 +61,7 @@ SPEAKER_LOCK = threading.Lock()
 SPEAKER_CONDITIONING: dict | None = None
 STATE = {
    "last_backend": None,
-    "xtts_failures": 0,
-    "piper_fallbacks": 0,
+    "qwen_failures": 0,
    "last_error": None,
 }

@@ -738,18 +735,6 @@ def synthesize_xtts(text: str, output_format: str,
    return _convert(_wav(_join_pcm(pcm_parts)), output_format, speed)


-def synthesize_piper(text: str, output_format: str,
-                     speed: float) -> tuple[bytes, str]:
-    upstream_format = "wav" if output_format == "pcm" else output_format
-    audio, content_type = _request(
-        f"{PIPER_URL}/tts",
-        payload={"text": text, "voice": "alloy", "speed": speed,
-                 "format": upstream_format},
-        timeout=PIPER_TIMEOUT,
-    )
-    return _convert(audio, "pcm", 1.0) if output_format == "pcm" else (audio, content_type)
-
-
 def synthesize_qwen(text: str, output_format: str,
                    speed: float) -> tuple[bytes, str]:
    text = prepare_for_qwen_speech(text)
@@ -814,22 +799,18 @@ def synthesize(text: str, output_format: str, speed: float) -> tuple[bytes, str]
                STATE["last_backend"] = "qwen3-tts-1.7b"
                STATE["last_error"] = None
            return audio
-        except Exception as exc:  # fallback must cover all Qwen failures
+        except Exception as exc:
            with STATE_LOCK:
-                STATE["xtts_failures"] += 1
+                STATE["qwen_failures"] += 1
                STATE["last_error"] = type(exc).__name__
+            raise
        finally:
            SYNTHESIS_LOCK.release()
    else:
        with STATE_LOCK:
-            STATE["xtts_failures"] += 1
+            STATE["qwen_failures"] += 1
            STATE["last_error"] = "queue-timeout"
-
-    audio = synthesize_piper(text, output_format, speed)
-    with STATE_LOCK:
-        STATE["last_backend"] = "piper"
-        STATE["piper_fallbacks"] += 1
-    return audio
+        raise RuntimeError("speech queue timeout")


 class Handler(BaseHTTPRequestHandler):
@@ -856,19 +837,20 @@ class Handler(BaseHTTPRequestHandler):
            self.send_json(HTTPStatus.NOT_FOUND, {"error": "not found"})
            return
        primary_ready = _reachable(QWEN_TTS_URL, "/health")
-        fallback_ready = _reachable(PIPER_URL, "/status")
        with STATE_LOCK:
            state = dict(STATE)
+        # This endpoint is also the container liveness check.  Qwen3-TTS is
+        # deliberately stopped in exclusive GPU modes such as Applio, so the
+        # gateway itself must stay healthy while reporting ready=false.
        self.send_json(
-            HTTPStatus.OK if fallback_ready else HTTPStatus.SERVICE_UNAVAILABLE,
+            HTTPStatus.OK,
            {
-                "ready": fallback_ready,
-                "engine": "qwen3-tts-with-piper-fallback",
+                "ready": primary_ready,
+                "engine": "qwen3-tts",
                "model": "Qwen3-TTS-12Hz-1.7B-Base",
                "voices": [VOICE_ALIAS],
                "speaker": QWEN_TTS_VOICE,
                "primary_ready": primary_ready,
-                "fallback_ready": fallback_ready,
                **state,
            },
        )
@@ -916,7 +898,7 @@ class Handler(BaseHTTPRequestHandler):
            with STATE_LOCK:
                STATE["last_error"] = type(exc).__name__
            self.send_json(HTTPStatus.SERVICE_UNAVAILABLE,
-                           {"error": "all local speech backends failed"})
+                           {"error": "local Qwen3-TTS backend failed"})
            return
        print(f"tts-gateway: synthesized via {STATE['last_backend']} in "
              f"{time.monotonic() - started:.2f}s")
@@ -992,5 +974,5 @@ class Handler(BaseHTTPRequestHandler):


 if __name__ == "__main__":
-    print(f"TTS gateway ready on {HOST}:{PORT}; primary={QWEN_TTS_VOICE}; fallback=Piper")
+    print(f"TTS gateway ready on {HOST}:{PORT}; backend=Qwen3-TTS; voice={QWEN_TTS_VOICE}")
    ThreadingHTTPServer((HOST, PORT), Handler).serve_forever()
@@ -76,7 +76,6 @@ start_proxy() {
 start_proxy 22   172.30.10.1:22
 start_proxy 8081 router:8081
 start_proxy 8085 tts-gateway:8085
-start_proxy 8091 piper:8085
 start_proxy 8099 llama-dashboard:8099
 start_proxy 7861 music-ui:3000
 start_proxy 7862 music-worker:7860
@@ -144,13 +144,12 @@ stt:
    model: "base"
    language: "de"

-# Reuse Athena's OpenAI-compatible TTS route. It currently serves XTTS v2 with
-# Annmarie Nele and transparently falls back to Piper when XTTS is unavailable.
+# Reuse Athena's OpenAI-compatible Qwen3-TTS route.
 tts:
  provider: "openai"
  speed: 1.0
  openai:
-    model: "piper"
+    model: "qwen3-tts"
    voice: "alloy"
    speed: 1.0
    base_url: "http://router:8081/v1"
@@ -25,7 +25,7 @@ image worker, then restores the previous LLM state.

 Only one heavy GPU path may be active. Do not manually start a second GPU
 worker around the controller. The lightweight dashboard, router, controller,
-gateway, UI, CPU-STT, Piper, backup and operator containers may remain active.
+gateway, UI, CPU-STT, backup and operator containers may remain active.

 ## Model profiles

@@ -33,7 +33,6 @@ gateway, UI, CPU-STT, Piper, backup and operator containers may remain active.
 - Medium: Qwen3.8-27B IQ4_XS-pure, 160,000 tokens, vision.
 - Large: the same Q4 model, 192,000 tokens, vision.
 - Ultra: the same Q4 model, 262,144 tokens, no vision projector.
- Beta 1: Qwen3.8-27B GSQ-RCO IQ3_S-MTP, 112,000 tokens; experimental only.
 - Uncensored: Abliterated Q4_K_M, 80,000 tokens, vision.

 Medium, Large, Ultra and Beta distribute their runtime across both GPUs. Do not
@@ -163,10 +163,10 @@ log "Tool-Container mit der neuen Stackdefinition aktivieren"
 log "Router und OpenWebUI mit den wiederhergestellten Schlüsseln neu erstellen"
 cd "$STACK_DIR"
 docker compose --env-file "$SECRETS_DIR/stack.env" up -d --force-recreate \
-  qwen3-tts piper tts-gateway router open-webui
+  qwen3-tts tts-gateway router open-webui

 deadline=$((SECONDS + 180))
-for container in mike-ai-qwen3-tts mike-ai-piper mike-ai-tts-gateway \
+for container in mike-ai-qwen3-tts mike-ai-tts-gateway \
  mike-ai-router mike-ai-open-webui; do
  until [[ $(docker inspect -f '{{if .State.Health}}{{.State.Health.Status}}{{else}}{{.State.Status}}{{end}}' \
      "$container" 2>/dev/null || true) == healthy ]]; do
@@ -83,7 +83,7 @@ with con:
    for key, value in (
        ("task.follow_up.enable", False),
        ("audio.tts.engine", "openai"),
-        ("audio.tts.model", "piper"),
+        ("audio.tts.model", "qwen3-tts"),
        ("audio.tts.voice", "alloy"),
        ("audio.tts.openai.api_base_url", "http://router:8081/v1"),
        ("web.search.enable", True),
@@ -7,9 +7,9 @@ SERVICE="${LLAMA_SERVICE:-mike-ai-llama-ui.service}"
 PROFILE="${1:-}"

 case "$PROFILE" in
-  fast|medium|beta1|large|ultra|uncensored) ;;
+  fast|medium|large|ultra|uncensored) ;;
  *)
-    echo "Usage: llama-profile {fast|medium|beta1|large|ultra|uncensored}" >&2
+    echo "Usage: llama-profile {fast|medium|large|ultra|uncensored}" >&2
    exit 2
    ;;
 esac
@@ -60,7 +60,7 @@ backup_root=$work/backup
 [[ -d $backup_root/volumes ]] || die "Backup enthält keine Docker-Volumes."

 # Stop only users of the restored volumes. WireGuard, SSH and networking stay up.
-for container in mike-ai-router mike-ai-profile-controller mike-ai-piper mike-ai-portainer; do
+for container in mike-ai-router mike-ai-profile-controller mike-ai-portainer; do
  if [[ $(docker inspect -f '{{.State.Running}}' "$container" 2>/dev/null || true) == true ]]; then
    docker stop "$container" >/dev/null
  fi
@@ -78,7 +78,6 @@ restore_volume() {
  rsync -a --delete "$source/" "$mountpoint/"
 }

-restore_volume mike-ai_piper-data "$backup_root/volumes/piper-data"
 restore_volume mike-ai_router-state "$backup_root/volumes/router-state"
 restore_volume mike-ai_router-images "$backup_root/volumes/router-images"
 restore_volume portainer_data "$backup_root/volumes/portainer-data"
@@ -2,21 +2,20 @@
 """AI Profile Router – OpenAI-kompatibler Proxy vor llama.cpp.

 Leitet OpenAI-kompatible Requests transparent an den lokalen llama.cpp-Server
-weiter (Streaming, Tool Calls, JSON) und schaltet zwischen sechs festen
+weiter (Streaming, Tool Calls, JSON) und schaltet zwischen fünf festen
 Profilen um:

    Profil   Kontext
    ------   --------
    fast     76800
    medium   160000
-    beta1    192000
    large    192000
    ultra    262144
    uncensored 80000

-Virtuelle Modelle:  qwen-fast, qwen-medium, qwen-beta-1, qwen-large,
+Virtuelle Modelle:  qwen-fast, qwen-medium, qwen-large,
                    qwen-ultra, qwen-uncensored
-Kommandos:          POST /fast, /medium, /beta1, /large, /ultra,
+Kommandos:          POST /fast, /medium, /large, /ultra,
                    /uncensored
                    GET  /status                 (Zustand)

@@ -26,7 +25,7 @@ Bildgenerierung und Editing (FLUX.2 Klein 9B FP8 beta):
                    GET  /images                 (Liste)
                    GET  /images/<datei>         (PNG-Download)

-Sprachausgabe (Qwen3-TTS auf RTX 3060, Piper als CPU-Fallback):
+Sprachausgabe (Qwen3-TTS auf RTX 3060):
                    POST /v1/audio/speech        (OpenAI-kompatibel)
                    GET  /v1/audio/voices        (verfügbare Stimmen)

@@ -177,15 +176,15 @@ IMAGE_QUALITY = {"standard": 4, "high": 4}
 IMAGE_DEFAULT_QUALITY = "standard"
 IMAGE_MAX_N = 4

-# --- Sprachausgabe (austauschbarer interner TTS-Worker, CPU-only) ---
+# --- Sprachausgabe (Qwen3-TTS über das interne Normalisierungs-Gateway) ---
 TTS_WORKER_URL = os.environ.get("TTS_WORKER_URL", "http://127.0.0.1:8085")
 TTS_TIMEOUT = float(os.environ.get("TTS_TIMEOUT", "300"))  # s, pro Synthese
 TTS_CONNECT_TIMEOUT = float(os.environ.get("TTS_CONNECT_TIMEOUT", "5"))
-TTS_MODEL = os.environ.get("TTS_MODEL", "xtts-v2")
+TTS_MODEL = os.environ.get("TTS_MODEL", "qwen3-tts")
 TTS_VOICES = tuple(v.strip() for v in os.environ.get(
-    "TTS_VOICES", "claribel").split(",") if v.strip())
+    "TTS_VOICES", "alloy").split(",") if v.strip())
 TTS_DEFAULT_VOICE = os.environ.get(
-    "TTS_DEFAULT_VOICE", TTS_VOICES[0] if TTS_VOICES else "claribel")
+    "TTS_DEFAULT_VOICE", TTS_VOICES[0] if TTS_VOICES else "alloy")
 TTS_FORMATS = ("mp3", "wav", "pcm")
 TTS_DEFAULT_FORMAT = "mp3"

@@ -2363,7 +2362,7 @@ class Handler(BaseHTTPRequestHandler):
        self.end_headers()
        self.wfile.write(data)

-    # ---------- Sprachausgabe (Qwen3-TTS mit Piper-Fallback) ----------
+    # ---------- Sprachausgabe (Qwen3-TTS) ----------

    def _speech(self) -> None:
        try:
@@ -2422,7 +2421,7 @@ class Handler(BaseHTTPRequestHandler):
                             "invalid_request_error", "invalid_speed")
            return

-        # Modell-Name optional; falls angegeben, muss es xtts-v2 sein.
+        # Modell-Name optional; falls angegeben, muss es Qwen3-TTS sein.
        model = data.get("model")
        if model is not None and model != TTS_MODEL:
            self._send_error(400, f"unbekanntes Modell: {model!r} "
@@ -2532,7 +2531,7 @@ class Handler(BaseHTTPRequestHandler):
            models.append({
                "id": TTS_MODEL,
                "object": "model",
-                "owned_by": "coqui-xtts",
+                "owned_by": "qwen",
                "type": "speech",
            })
        return {"object": "list", "data": models}
@@ -8,10 +8,6 @@
      "context": 160000,
      "model_alias": "qwen-medium"
    },
-    "beta1": {
-      "context": 112000,
-      "model_alias": "qwen-beta-1"
-    },
    "large": {
      "context": 192000,
      "model_alias": "qwen-large"
@@ -35,7 +35,7 @@ def main() -> int:
    speech = request(
        "/audio/speech",
        json.dumps({
-            "model": "piper",
+            "model": "qwen3-tts",
            "voice": "alloy",
            "response_format": "wav",
            "input": TEST_TEXT,
@@ -27,7 +27,6 @@ for name in \
  mike-ai-wireguard-gateway \
  mike-ai-profile-controller \
  mike-ai-router \
-  mike-ai-piper \
  mike-ai-qwen3-tts \
  mike-ai-tts-gateway \
  mike-ai-llama-dashboard \
@@ -45,7 +44,7 @@ done
 pass "Dashboard und Portainer besitzen stabile Netzwerk-Namespaces"

 active_llama=$(docker ps --format '{{.Names}}' | \
-  grep -Ec '^mike-ai-llama-(fast|medium|beta1|large|ultra|uncensored)$' || true)
+  grep -Ec '^mike-ai-llama-(fast|medium|large|ultra|uncensored)$' || true)
 [[ $active_llama -eq 1 ]] || fail "$active_llama aktive llama-Profile (erwartet: 1)"
 pass "Genau ein llama.cpp-Profil ist aktiv"