[NEW][audiobreezettsv3][1.0.17]

audiobreezettsv3/Chart.yaml

@@ -0,0 +1,6 @@
+apiVersion: v2
+appVersion: "1.0.2"
+description: Breeze TTS 2 text to speech with built-in voices, cloning and design.
+name: audiobreezettsv3
+type: application
+version: 1.0.17

audiobreezettsv3/OlaresManifest.yaml

@@ -0,0 +1,252 @@
+olaresManifest.version: "0.12.0"
+olaresManifest.type: app
+apiVersion: 'v3'
+metadata:
+  name: audiobreezettsv3
+  icon: https://app.cdn.olares.com/appstore/audioengines/soulx/icon.png
+  description: "A text-to-speech model with built-in, cloned, and description-based voices"
+  appid: audiobreezettsv3
+  title: Breeze TTS 2
+  version: '1.0.17'
+  categories:
+    - AI
+    - models
+  tags:
+  - text-to-speech
+  - audio
+
+# Internal shared API stub: in-cluster callers use http://sharedentrances-api.<ns>:80 + /v1.
+sharedEntrances:
+  - name: audiobreezettsv3api
+    host: sharedentrances-api
+    port: 0
+    title: Breeze TTS 2 API
+    invisible: true
+    authLevel: internal
+    icon: https://app.cdn.olares.com/appstore/audioengines/soulx/icon.png
+# Entrance -> llm-init (download-svc:8090): dashboard on "/", engine /v1/* reverse-proxied; authLevel internal.
+entrances:
+  - name: audiobreezettsv3
+    port: 8090
+    host: download-svc
+    title: Breeze TTS 2
+    icon: https://app.cdn.olares.com/appstore/audioengines/soulx/icon.png
+    openMethod: window
+    authLevel: internal
+workloadReplicas:
+  audiobreezettsv3: 1
+  llminit: 1
+spec:
+  onlyAdmin: true
+  versionName: 'BreezeBlue/Breeze-TTS-2'
+  fullDescription: |
+    Breeze TTS 2 turns written Chinese or English text into spoken audio. You can use a built-in voice, copy a voice from a reference recording, or create a new voice by describing how it should sound.
+
+    **Key features**
+
+    - Generate speech in Chinese and English.
+    - Choose from the built-in voices available in this app.
+    - Clone a voice from clean reference audio and its matching transcript.
+    - Create a voice from a natural-language description without reference audio.
+    - Save and reuse cloned or created voices on the same Olares installation.
+
+    A compatible NVIDIA GPU with at least `12Gi` of GPU memory is required.
+
+    **Usage notes**
+
+    Breeze TTS 2 model weights, derivative models, and self-hosted outputs are licensed for research and non-commercial use only. Users must have the necessary rights and consent for reference recordings, cloned voices, and generated audio.
+
+    See the [Breeze TTS 2 model card](https://huggingface.co/BreezeBlue/Breeze-TTS-2) for license details and responsible-use requirements.
+  developer: BreezeBlue
+  website: https://huggingface.co/BreezeBlue/Breeze-TTS-2
+  sourceCode: https://github.com/breezeblue-ai/breeze-tts
+  submitter: Olares
+  locale:
+    - en-US
+    - zh-CN
+    - fr-FR
+    - de-DE
+    - ja-JP
+    - it-IT
+    - es-ES
+  doc: https://huggingface.co/BreezeBlue/Breeze-TTS-2
+  license:
+    - text: Apache-2.0
+      url: https://github.com/breezeblue-ai/breeze-tts/blob/main/LICENSE
+    - text: BreezeBlue Research / Non-Commercial (model weights)
+      url: https://huggingface.co/BreezeBlue/Breeze-TTS-2/blob/main/LICENSE
+
+  accelerator:
+    # Sum rule: accelerator >= engine + llm-init (200m/256Mi req, 1/4Gi lim). On nvidia-gb10 the
+    # engine's memory request AND limit become the GPU quota (unified memory).
+    - mode: nvidia
+      limitedCpu: "7"
+      requiredCpu: "700m"
+      requiredDisk: 50Mi
+      limitedDisk: 200Gi
+      limitedMemory: "20Gi"
+      requiredMemory: "4352Mi"
+      requiredGPUMemory: "12Gi"
+      limitedGPUMemory: "12Gi"
+    - mode: nvidia-gb10
+      limitedCpu: "7"
+      requiredCpu: "700m"
+      requiredDisk: 50Mi
+      limitedDisk: 200Gi
+      limitedMemory: "16Gi"
+      requiredMemory: "12544Mi"
+
+  supportArch:
+    - amd64
+    - arm64
+permission:
+  appData: true
+  appCache: true
+  appCommon: true
+options:
+  apiTimeout: 0
+  LLMGatewaySupported: true
+  shared: true
+  templateOnly: false
+  dependencies:
+    - name: olares
+      version: '>=1.12.7-0'
+      type: system
+envs:
+  - envName: MODEL_SOURCE
+    required: true
+    type: string
+    editable: false
+    applyOnChange: false
+    default: "hf://BreezeBlue/Breeze-TTS-2"
+    description: "HuggingFace source llm-init downloads into the shared cache."
+    regex: "^hf://"
+  - envName: MODEL_NAME
+    required: true
+    type: string
+    editable: false
+    applyOnChange: false
+    default: "BreezeBlue/Breeze-TTS-2"
+    description: "Model id advertised to clients; must match the engine /v1/models id."
+  - envName: MODEL_MODE
+    required: true
+    type: string
+    editable: false
+    applyOnChange: false
+    default: "tts"
+    options:
+    - title: TTS
+      value: tts
+    description: "Speech synthesis is its own mode: it answers the ElevenLabs-shaped voice routes recognition engines do not have, and Router routes by mode. Requires an Olares whose Router and Model Console know `tts`."
+  - envName: MODEL_SUPPORTS
+    required: true
+    type: string
+    editable: false
+    applyOnChange: false
+    default: "tts,tts_clone,tts_design"
+    multiSelect: true
+    splitter: ","
+    options:
+    - title: TTS (voices + speech)
+      value: tts
+    - title: TTS clone (save a reference as a voice_id)
+      value: tts_clone
+    - title: TTS design (build a voice from a written description)
+      value: tts_design
+    description: "One 3B checkpoint covers all three. `tts` mounts list/speak; `tts_design` is advertised on the model card for `/v1/text-to-voice/*`; `tts_clone` mounts POST /v1/voices/add."
+  - envName: TTS_SPEED_MIN
+    required: true
+    type: string
+    editable: false
+    applyOnChange: false
+    default: "0.25"
+    description: "Minimum synthesis speed published in the model card."
+  - envName: TTS_SPEED_DEFAULT
+    required: true
+    type: string
+    editable: false
+    applyOnChange: false
+    default: "1.0"
+    description: "Normal synthesis speed published in the model card."
+  - envName: TTS_SPEED_MAX
+    required: true
+    type: string
+    editable: false
+    applyOnChange: false
+    default: "4.0"
+    description: "Maximum synthesis speed published in the model card."
+  - envName: REQUIRED_GPU_MEMORY
+    required: true
+    type: string
+    editable: false
+    applyOnChange: false
+    default: "12Gi"
+    description: "GPU memory required for this model. CUDA only; `0` is not supported."
+  - envName: CPU_REQUEST
+    required: true
+    type: string
+    editable: false
+    applyOnChange: false
+    default: "500m"
+  - envName: MEMORY_REQUEST
+    required: true
+    type: string
+    editable: false
+    applyOnChange: false
+    default: "4Gi"
+  - envName: CPU_LIMIT
+    required: true
+    type: string
+    editable: false
+    applyOnChange: false
+    default: "6"
+  - envName: MEMORY_LIMIT
+    required: true
+    type: string
+    editable: false
+    applyOnChange: false
+    default: "16Gi"
+  - envName: LOG_LEVEL
+    required: false
+    type: string
+    editable: true
+    applyOnChange: true
+    default: "info"
+    options:
+    - title: Debug
+      value: debug
+    - title: Info
+      value: info
+    - title: Warn
+      value: warn
+    - title: Error
+      value: error
+  - envName: ENGINE_IMAGE
+    required: false
+    type: string
+    editable: true
+    applyOnChange: true
+  - envName: ENGINE_ARGS
+    required: false
+    type: string
+    editable: true
+    applyOnChange: true
+    default: "--cfg-scale 1 --design-cfg 4 --ref-min-seconds 3 --ref-max-seconds 30 --ref-max-mb 5 --design-min-chars 3 --design-max-chars 500 --max-new-tokens 1500 --max-seq-len 2048"
+    description: "Engine-native flags (llm-init ENGINE_ARGS). Filling this replaces the whole
+      string. --cfg-scale clone CFG (official 1). --design-cfg design/direction CFG (official 4).
+      --ref-min-seconds / --ref-max-seconds / --ref-max-mb reject at /v1/voices/add (0 = off).
+      --design-min-chars / --design-max-chars reject at design (0 = off). --max-new-tokens and
+      --max-seq-len are official infer.py windows, and they also set how much text one slice
+      can hold, so lowering them makes the slices shorter. Also: --seed, --voice-dir,
+      --preview-text, --attn-implementation. Do not pass --fast-all: the CUDA Graph path needs
+      a C compiler this image does not carry, so the engine fails to load."
+  - envName: HF_ENDPOINT
+    required: false
+    applyOnChange: true
+    valueFrom:
+      envName: OLARES_SYSTEM_HUGGINGFACE_SERVICE
+  - envName: HF_TOKEN
+    required: false
+    applyOnChange: true
+    valueFrom:
+      envName: OLARES_SYSTEM_HUGGINGFACE_TOKEN

audiobreezettsv3/i18n/de-DE/OlaresManifest.yaml

@@ -0,0 +1,23 @@
+metadata:
+  title: Breeze TTS 2
+  description: Ein Text-zu-Sprache-Modell mit eingebauten, geklonten und per Beschreibung erstellten Stimmen
+
+spec:
+  fullDescription: |
+    Breeze TTS 2 wandelt geschriebenen chinesischen oder englischen Text in Sprache um. Es kann eine eingebaute Stimme verwendet, eine Stimme aus einer Referenzaufnahme kopiert oder eine neue Stimme durch eine Beschreibung des Klangs erstellt werden.
+
+    **Hauptfunktionen**
+
+    - Sprache auf Chinesisch und Englisch erzeugen.
+    - Aus den in dieser App verfügbaren eingebauten Stimmen wählen.
+    - Eine Stimme aus einer sauberen Referenzaufnahme und dem zugehörigen Transkript klonen.
+    - Eine Stimme aus einer natürlichsprachlichen Beschreibung ohne Referenzaufnahme erstellen.
+    - Geklonte oder erstellte Stimmen auf derselben Olares-Installation speichern und erneut verwenden.
+
+    Eine kompatible NVIDIA-GPU mit mindestens `12Gi` GPU-Speicher ist erforderlich.
+
+    **Nutzungshinweise**
+
+    Die Modellgewichte von Breeze TTS 2, abgeleitete Modelle und selbst gehostete Ausgaben sind nur für Forschung und nicht kommerzielle Nutzung lizenziert. Für Referenzaufnahmen, geklonte Stimmen und erzeugtes Audio müssen die erforderlichen Rechte und Einwilligungen vorliegen.
+
+    Details zu Lizenz und verantwortungsvoller Nutzung stehen auf der [Breeze TTS 2 Modellkarte](https://huggingface.co/BreezeBlue/Breeze-TTS-2).

audiobreezettsv3/i18n/en-US/OlaresManifest.yaml

@@ -0,0 +1,23 @@
+metadata:
+  title: Breeze TTS 2
+  description: A text-to-speech model with built-in, cloned, and description-based voices
+
+spec:
+  fullDescription: |
+    Breeze TTS 2 turns written Chinese or English text into spoken audio. You can use a built-in voice, copy a voice from a reference recording, or create a new voice by describing how it should sound.
+
+    **Key features**
+
+    - Generate speech in Chinese and English.
+    - Choose from the built-in voices available in this app.
+    - Clone a voice from clean reference audio and its matching transcript.
+    - Create a voice from a natural-language description without reference audio.
+    - Save and reuse cloned or created voices on the same Olares installation.
+
+    A compatible NVIDIA GPU with at least `12Gi` of GPU memory is required.
+
+    **Usage notes**
+
+    Breeze TTS 2 model weights, derivative models, and self-hosted outputs are licensed for research and non-commercial use only. Users must have the necessary rights and consent for reference recordings, cloned voices, and generated audio.
+
+    See the [Breeze TTS 2 model card](https://huggingface.co/BreezeBlue/Breeze-TTS-2) for license details and responsible-use requirements.

audiobreezettsv3/i18n/es-ES/OlaresManifest.yaml

@@ -0,0 +1,23 @@
+metadata:
+  title: Breeze TTS 2
+  description: Un modelo de síntesis de voz con voces integradas, clonadas y basadas en una descripción
+
+spec:
+  fullDescription: |
+    Breeze TTS 2 convierte texto chino o inglés en voz. Se puede usar una voz integrada, copiar una voz a partir de una grabación de referencia o crear una voz nueva describiendo cómo debe sonar.
+
+    **Funciones principales**
+
+    - Generar voz en chino e inglés.
+    - Elegir entre las voces integradas disponibles en esta aplicación.
+    - Clonar una voz a partir de un audio de referencia nítido y su transcripción.
+    - Crear una voz a partir de una descripción en lenguaje natural, sin audio de referencia.
+    - Guardar y reutilizar voces clonadas o creadas en la misma instalación de Olares.
+
+    Se requiere una GPU NVIDIA compatible con al menos `12Gi` de memoria GPU.
+
+    **Notas de uso**
+
+    Los pesos del modelo Breeze TTS 2, los modelos derivados y las salidas de un alojamiento propio están licenciados solo para investigación y uso no comercial. Para las grabaciones de referencia, las voces clonadas y el audio generado deben existir los derechos y el consentimiento necesarios.
+
+    Consulte la [ficha del modelo Breeze TTS 2](https://huggingface.co/BreezeBlue/Breeze-TTS-2) para los detalles de licencia y los requisitos de uso responsable.

audiobreezettsv3/i18n/fr-FR/OlaresManifest.yaml

@@ -0,0 +1,23 @@
+metadata:
+  title: Breeze TTS 2
+  description: Un modèle de synthèse vocale avec voix intégrées, clonées et décrites
+
+spec:
+  fullDescription: |
+    Breeze TTS 2 transforme du texte chinois ou anglais en parole. On peut utiliser une voix intégrée, copier une voix à partir d'un enregistrement de référence, ou créer une voix en décrivant le son souhaité.
+
+    **Fonctionnalités principales**
+
+    - Générer de la parole en chinois et en anglais.
+    - Choisir parmi les voix intégrées fournies par cette application.
+    - Cloner une voix à partir d'un enregistrement de référence net et de sa transcription.
+    - Créer une voix à partir d'une description en langage naturel, sans enregistrement de référence.
+    - Enregistrer et réutiliser les voix clonées ou créées sur la même installation Olares.
+
+    Un GPU NVIDIA compatible avec au moins `12Gi` de mémoire GPU est requis.
+
+    **Notes d'utilisation**
+
+    Les poids du modèle Breeze TTS 2, les modèles dérivés et les sorties d'un hébergement autonome sont concédés sous licence uniquement pour la recherche et un usage non commercial. Les utilisateurs doivent disposer des droits et du consentement nécessaires pour les enregistrements de référence, les voix clonées et l'audio généré.
+
+    Voir la [fiche du modèle Breeze TTS 2](https://huggingface.co/BreezeBlue/Breeze-TTS-2) pour les détails de licence et les exigences d'usage responsable.

audiobreezettsv3/i18n/it-IT/OlaresManifest.yaml

@@ -0,0 +1,23 @@
+metadata:
+  title: Breeze TTS 2
+  description: Un modello di sintesi vocale con voci integrate, clonate e basate su una descrizione
+
+spec:
+  fullDescription: |
+    Breeze TTS 2 trasforma testo cinese o inglese in parlato. Si può usare una voce integrata, copiare una voce da una registrazione di riferimento oppure crearne una nuova descrivendo come deve suonare.
+
+    **Funzionalità principali**
+
+    - Generare parlato in cinese e in inglese.
+    - Scegliere tra le voci integrate disponibili in questa app.
+    - Clonare una voce da un audio di riferimento pulito e dalla relativa trascrizione.
+    - Creare una voce da una descrizione in linguaggio naturale, senza audio di riferimento.
+    - Salvare e riutilizzare le voci clonate o create sulla stessa installazione Olares.
+
+    È necessaria una GPU NVIDIA compatibile con almeno `12Gi` di memoria GPU.
+
+    **Note sull'uso**
+
+    I pesi del modello Breeze TTS 2, i modelli derivati e gli output di un'installazione self-hosted sono concessi in licenza solo per ricerca e uso non commerciale. Per le registrazioni di riferimento, le voci clonate e l'audio generato occorrono i diritti e il consenso necessari.
+
+    Vedere la [scheda del modello Breeze TTS 2](https://huggingface.co/BreezeBlue/Breeze-TTS-2) per i dettagli di licenza e i requisiti di uso responsabile.