Custom video avatar speech synthesis fails with personalVoiceName access/not available error in Foundry

Hello @Marc Teipel ,

Welcome to Microsoft Q&A .Thank you for reaching out to us.

Based on the information available so far, the behavior appears to be isolated to the speech synthesis stage when the custom avatar attempts to use its included voice, rather than a general Azure AI Foundry Agent, Speech service connectivity, or RBAC issue. Agent responses are generated successfully, and standard Azure AI Speech voices synthesize correctly, which indicates that the overall Agent workflow, Speech service connectivity, and avatar rendering pipeline are functioning as expected. The failure occurs specifically when the avatar's included voice is selected and the runtime returns:"Speech synthesis failed: you don't have access to this personalVoiceName or it's not available."

The screenshot shows that the selected custom avatar contains a"Voice included" badge and that"Use avatar's included voice" is enabled. In Azure AI Speech, avatar experiences can use different voice technologies, including:

• Voice Sync for Avatar – trained together with a custom video avatar and associated with that avatar experience.
• Personal Voice – a separately created voice profile.
• Professional Voice (Custom Neural Voice) – a separately trained custom voice endpoint.

Because these voice types follow different deployment and configuration models, the current behavior suggests that the issue may be related to how the avatar-associated voice is being resolved during synthesis rather than a general Speech service failure.

Please check if the following steps help-

1. Confirming whether the issue is isolated to the avatar's included voice Please perform the following validation:
   1. Open the Avatar section.
   2. Uncheck"Use avatar's included voice".
   3. Select a standard Azure AI Speech voice such as en-US-AvaNeural.
   4. Run the conversation again.
   Expected result
   • If speech synthesis and avatar rendering complete successfully, this confirms that the Agent runtime, Speech service, and avatar pipeline are healthy, and that the issue is isolated to the avatar-associated voice.
2. Verifying Speech resource alignment One of the more common causes of this behavior is a mismatch between the Speech resource used during avatar creation and the Speech resource used during inference. Please confirm:
   • Which Azure AI Speech resource was used when the custom avatar was created.
   • Whether the Azure AI Foundry project is currently configured to use that same Speech resource.
3. Verifying region alignment Please confirm the regions for:
   • Azure AI Foundry Project
   • Azure AI Speech Resource
   • Custom Video Avatar
   These resources should be deployed in compatible supported regions. Region misalignment can prevent avatar-associated voice assets from being resolved successfully during synthesis.
4. Validating the voice outside Foundry To determine whether the issue is specific to Foundry or the underlying voice asset, please verify whether the same voice can be synthesized successfully through:
   • Azure AI Speech Studio
     • Azure AI Speech SDK
       • Azure AI Speech REST API
   If the voice cannot be synthesized outside Foundry, the issue is more likely related to voice availability, deployment, entitlement, or resource configuration. If the voice works outside Foundry but fails only through Agent Voice Mode, the issue may be isolated to the Foundry integration path.
5. Verifying voice availability and deployment status Please confirm whether any of the following occurred after avatar creation:
   • The avatar was retrained.
     • The associated voice was recreated.
       • The Speech resource was changed.
         • Resources were moved between subscriptions or regions.
   Additionally, please verify that:
   • The custom avatar and associated voice are available within the serving Speech resource.
   • The avatar assets are not only present in the original training project.
   • The required avatar-related capabilities remain available for the serving resource and region
   1. Validating runtime identity access
   Another area worth validating is the identity used during runtime. Even when user-level RBAC permissions are correctly assigned, the Foundry runtime may access Speech services through a managed identity or service identity. Please verify that the runtime identity has access to the same Speech resource and configuration associated with the avatar voice.

The following references might be helpful , please check them out

• Custom text to speech avatar overview - Speech service - Foundry Tools | Microsoft Learn
• Voice sync for avatar - Speech service - Azure AI services | Microsoft Learn
• Azure Speech in Foundry Tools known issues - Foundry Tools | Microsoft Learn
• Limited Access features for Foundry Tools - Foundry Tools | Microsoft Learn

Please let us know if the response was helpful

Thank you