charts

voicestudio

VoiceStudio (formerly OmniVoice Studio) — an open-source, fully local ElevenLabs alternative: voice cloning, voice design, dubbing, dictation, transcription and audiobook creation. This Helm chart runs the headless server image (API + web UI on a single port), CPU by default with optional NVIDIA or AMD ROCm acceleration, driven from values.yaml via the bjw-s common library.

TL;DR

helm repo add obeone https://charts.obeone.cloud
helm repo update
helm install voicestudio obeone/voicestudio

About

VoiceStudio (formerly OmniVoice Studio) is an open-source, fully local alternative to ElevenLabs: voice cloning, voice design, video dubbing, dictation, transcription and audiobook creation. This chart runs its headless server image, where one port serves both the API and the web UI, on CPU by default with optional NVIDIA or AMD ROCm acceleration.

Application: github.com/debpalash/VoiceStudio
Container image: ghcr.io/debpalash/voicestudio
Chart source: charts/voicestudio

Prerequisites

Helm 3
A Kubernetes cluster matching the chart’s kubeVersion constraint (see Chart.yaml)
linux/amd64 nodes: upstream publishes no arm64 image
A Secret holding the administrator key (see below)

Configuration

This chart is built on the bjw-s-labs common library. Most configuration keys (controllers, service, ingress, persistence, …) follow its schema; see the common library documentation for everything it supports beyond what is spelled out in values.yaml.

Defaults are meant to work out of the box on any cluster. The full list of options lives in values.yaml, is validated by values.schema.json at install time, and is documented in the Values section below. Override it with your own values file:

helm install voicestudio obeone/voicestudio -f my-values.yaml

Administrator key

VoiceStudio requires an administrator API key, and the web UI asks for it on first use. The pod reads it from the Secret <fullname>-api-key (key api-key) and waits until it exists. Create it before or right after installing:

kubectl create secret generic voicestudio-api-key \
  --from-literal=api-key="$(python3 -c 'import secrets; print(secrets.token_urlsafe(32))')"

Alternatively set secrets.api-key.enabled=true and pass the value with --set-string secrets.api-key.stringData.api-key=....

GPU acceleration

The default image is the CUDA build, which falls back to CPU when no GPU is visible. For an NVIDIA GPU, request it in both requests and limits, and set the RuntimeClass if your cluster needs one:

defaultPodOptions:
  runtimeClassName: nvidia
controllers:
  main:
    containers:
      main:
        resources:
          requests:
            nvidia.com/gpu: 1
          limits:
            nvidia.com/gpu: 1

For an AMD GPU, switch to the ROCm image (image.tag: "0.5.6-rocm") and request amd.com/gpu through the AMD device plugin.

Image pull fails with a digest mismatch

The image carries a single PyTorch layer of about 4 GB. If the pull fails with unexpected commit digest (the download was cut short), switch to the Docker Hub mirror, which publishes the same image with identical tags:

controllers:
  main:
    containers:
      main:
        image:
          repository: docker.io/palashdeb/omnivoice-studio

Upgrading

helm repo update
helm upgrade voicestudio obeone/voicestudio

Each release lists its changes in the Artifact Hub changelog; give it a look before jumping across several chart versions.

Uninstalling

helm uninstall voicestudio

PersistentVolumeClaims created by the chart are kept around: delete them manually if you also want the data gone.

Requirements

Kubernetes: >=1.31.0-0

Repository	Name	Version
https://bjw-s-labs.github.io/helm-charts	common	5.2.1

Values

Key	Type	Default	Description
controllers.main.containers.main.env	object	`{"HF_HOME":"/app/omnivoice_data/huggingface","OMNIVOICE_API_KEY":{"valueFrom":{"secretKeyRef":{"key":"api-key","name":"-api-key"}}},"OMNIVOICE_BIND_HOST":"0.0.0.0","OMNIVOICE_DATA_DIR":"/app/omnivoice_data","OMNIVOICE_SERVER_MODE":"1","PYTHONUNBUFFERED":"1"}`	Environment of the VoiceStudio backend.
controllers.main.containers.main.env.OMNIVOICE_API_KEY	object	`{"valueFrom":{"secretKeyRef":{"key":"api-key","name":"-api-key"}}}`	Administrator API key, REQUIRED: settings, diagnostics and other admin actions are refused without it, and the web UI asks for it (the upstream Compose file will not start without one). Read from the `<fullname>-api-key` Secret, which you either create yourself (see NOTES) or let the chart create via `secrets.api-key` below. The pod stays in CreateContainerConfigError until it exists.
controllers.main.containers.main.image.pullPolicy	string	`"IfNotPresent"`
controllers.main.containers.main.image.repository	string	`"ghcr.io/debpalash/voicestudio"`	Container image. Mirrored on Docker Hub as `palashdeb/omnivoice-studio` with identical tags.
controllers.main.containers.main.image.tag	string	`""`	Image tag. Defaults to the chart’s appVersion (CUDA build, which also runs CPU-only when no NVIDIA GPU is present). For AMD GPUs use the ROCm variant, e.g. “0.5.6-rocm”.
controllers.main.containers.main.probes.liveness.custom	bool	`true`
controllers.main.containers.main.probes.liveness.enabled	bool	`true`
controllers.main.containers.main.probes.liveness.spec.failureThreshold	int	`5`
controllers.main.containers.main.probes.liveness.spec.httpGet.path	string	`"/health"`
controllers.main.containers.main.probes.liveness.spec.httpGet.port	int	`3900`
controllers.main.containers.main.probes.liveness.spec.periodSeconds	int	`30`
controllers.main.containers.main.probes.liveness.spec.timeoutSeconds	int	`10`
controllers.main.containers.main.probes.readiness.custom	bool	`true`
controllers.main.containers.main.probes.readiness.enabled	bool	`true`
controllers.main.containers.main.probes.readiness.spec.failureThreshold	int	`3`
controllers.main.containers.main.probes.readiness.spec.httpGet.path	string	`"/health"`
controllers.main.containers.main.probes.readiness.spec.httpGet.port	int	`3900`
controllers.main.containers.main.probes.readiness.spec.periodSeconds	int	`10`
controllers.main.containers.main.probes.readiness.spec.timeoutSeconds	int	`5`
controllers.main.containers.main.probes.startup.custom	bool	`true`
controllers.main.containers.main.probes.startup.enabled	bool	`true`
controllers.main.containers.main.probes.startup.spec.failureThreshold	int	`90`
controllers.main.containers.main.probes.startup.spec.httpGet.path	string	`"/health"`
controllers.main.containers.main.probes.startup.spec.httpGet.port	int	`3900`
controllers.main.containers.main.probes.startup.spec.periodSeconds	int	`10`
controllers.main.containers.main.probes.startup.spec.timeoutSeconds	int	`5`
controllers.main.containers.main.resources	object	`{"limits":{"memory":"8Gi"},"requests":{"cpu":"500m","memory":"2Gi"}}`	Resource envelope. Speech models are memory-hungry; no CPU limit so inference can burst. To use an NVIDIA GPU, add `nvidia.com/gpu: 1` to BOTH requests and limits (and set runtimeClassName if your cluster needs it). For AMD, request `amd.com/gpu: 1` with the ROCm image.
controllers.main.containers.main.securityContext.allowPrivilegeEscalation	bool	`false`
controllers.main.containers.main.securityContext.capabilities.drop[0]	string	`"ALL"`
controllers.main.pod.labels	object	`{}`	Extra pod labels. Example for Sablier scale-to-zero on a shared GPU:
controllers.main.strategy	string	`"Recreate"`
controllers.main.type	string	`"deployment"`
defaultPodOptions	object	`{"automountServiceAccountToken":false,"nodeSelector":{"kubernetes.io/arch":"amd64"},"runtimeClassName":"","securityContext":{"seccompProfile":{"type":"RuntimeDefault"}}}`	Pod-wide options applied to every controller in this chart.
defaultPodOptions.nodeSelector	object	`{"kubernetes.io/arch":"amd64"}`	Node selection. Upstream only publishes linux/amd64 images (CUDA and ROCm alike), so the pod is pinned to amd64 nodes. Add a GPU host label here when you enable acceleration.
defaultPodOptions.runtimeClassName	string	`""`	RuntimeClass exposing the GPU to the pod (e.g. “nvidia”). Leave empty for CPU-only clusters, or for AMD GPUs exposed through a device plugin.
ingress	object	`{"main":{"annotations":{},"className":"","enabled":false,"hosts":[{"host":"chart-example.local","paths":[{"path":"/","pathType":"Prefix","service":{"identifier":"main","port":"http"}}]}],"tls":[{"hosts":["chart-example.local"],"secretName":"tls-chart-example-local"}]}}`	Ingress. Disabled by default; flip enabled and set a real host to expose the UI. The backend speaks plain HTTP and the admin key travels with requests, so terminate TLS at the ingress and never expose it unencrypted.
persistence	object	`{"data":{"accessMode":"ReadWriteOnce","enabled":true,"globalMounts":[{"path":"/app/omnivoice_data"}],"size":"20Gi","type":"persistentVolumeClaim"},"dshm":{"globalMounts":[{"path":"/dev/shm"}],"medium":"Memory","sizeLimit":"1Gi","type":"emptyDir"}}`	Storage.
persistence.data	object	`{"accessMode":"ReadWriteOnce","enabled":true,"globalMounts":[{"path":"/app/omnivoice_data"}],"size":"20Gi","type":"persistentVolumeClaim"}`	Everything VoiceStudio keeps: SQLite database, user voices, settings and the Hugging Face model cache (~4 GB on first run, more as you add engines).
persistence.data.globalMounts	list	`[{"path":"/app/omnivoice_data"}]`	Reuse a pre-created PVC instead. existingClaim: “”
route	object	`{"main":{"enabled":false,"hostnames":["chart-example.local"],"kind":"HTTPRoute","parentRefs":[{"name":"gateway","namespace":"gateway-system"}],"rules":[{"backendRefs":[{"identifier":"main","port":3900}],"matches":[{"path":{"type":"PathPrefix","value":"/"}}]}]}}`	Gateway API HTTPRoute, mirroring the Ingress above. Disabled by default: pick either Ingress or HTTPRoute, not both. Requires the Gateway API CRDs and an existing Gateway in the cluster.
route.main.enabled	bool	`false`	Enable the HTTPRoute. Mutually exclusive with `ingress.main.enabled`.
route.main.hostnames	list	`["chart-example.local"]`	Hostnames served by this route.
route.main.kind	string	`"HTTPRoute"`	Route kind. HTTPRoute, GRPCRoute, TCPRoute, TLSRoute or UDPRoute.
route.main.parentRefs	list	`[{"name":"gateway","namespace":"gateway-system"}]`	Gateways this route attaches to.
route.main.rules	list	`[{"backendRefs":[{"identifier":"main","port":3900}],"matches":[{"path":{"type":"PathPrefix","value":"/"}}]}]`	Routing rules. `identifier` refers to a Service defined above.
secrets	object	`{"api-key":{"enabled":false,"stringData":{"api-key":""},"suffix":"api-key"}}`	Secrets managed by the chart.
secrets.api-key	object	`{"enabled":false,"stringData":{"api-key":""},"suffix":"api-key"}`	Chart-managed administrator key, rendered as `<fullname>-api-key`. Disabled by default so the key never lands in a values file by accident: create the Secret out of band (see NOTES). If you enable it, pass the value at install time (`--set-string secrets.api-key.stringData.api-key=...`).
service	object	`{"main":{"controller":"main","ports":{"http":{"port":3900,"protocol":"TCP","targetPort":3900},"worker":{"enabled":false,"port":7443,"protocol":"TCP","targetPort":7443}},"type":"ClusterIP"}}`	Service exposing the API and web UI inside the cluster.
service.main.ports.worker	object	`{"enabled":false,"port":7443,"protocol":"TCP","targetPort":7443}`	TLS control plane for remote GPU workers. Disabled by default; enable it together with OMNIVOICE_WORKER_PORT only if you enroll workers.

Verifying the chart signature

Charts in this repository are signed with GPG and every release ships a provenance file. The public key is available at charts.obeone.cloud/public_key.gpg, fingerprint B9FE852F28888D27F8C9A11CD33E04CD22E335CE.

# Import the signing key into a legacy keyring (helm verifies with GnuPG v1 keyrings)
curl -fsSL https://charts.obeone.cloud/public_key.gpg | gpg --import
gpg --export > ~/.gnupg/pubring.gpg

# Pull the chart and check it against its provenance file
helm pull --verify obeone/voicestudio

Support

This is a personal chart repository, maintained on a best-effort basis. Bug reports and contributions are welcome on GitHub.

Autogenerated from chart metadata using helm-docs v1.14.2