A publisher uses the Speech batch synthesis API to create audiobook chapters from SSML.

How it works

Speech batch synthesis turns large amounts of text or SSML into audio asynchronously. Where the results go depends on destinationContainerUrl:

• If you set it (a container URL with a SAS token), results are written to your writable Azure container. SAS with stored access policies isn't supported.
• If you don't, the Speech service stores the results in a container managed by Microsoft, and deleting the job deletes the result data too.

The batch synthesis pages are explicit about your own container: if you specify a destination storage account, it must allow access from all networks. The service doesn't support network-restricted storage accounts. A "selected networks" account is network-restricted, which is why delivery fails. The fix is a destination account that allows access from all networks, or omit the destination so results go to the Microsoft-managed container.

The other options, and when you'd use them

• Grant the Speech managed identity Storage Blob Data Contributor: The limitation is about network restriction, not permissions. Network-restricted accounts aren't supported.
• Switch inputKind to PlainText without a voice: The input format doesn't change where audio is written, and with PlainText you must also set the voice in synthesisConfig, so the request would be incomplete.
• Private endpoint to the restricted account: The account would still be network-restricted, which the service doesn't support.

Exam tip

Batch synthesis destination storage = must allow all networks. No destination = Microsoft-managed container.

References

• Batch synthesis API for text to speech
• Batch synthesis properties for text to speech