Speech to Text (STT) & Azure Cognitive Services Data Retention

How audio is passed to Azure Cognitive Services and whether it is retained – General Overview 

• Audio Transmission:
   When you use Azure Cognitive Speech Services (Speech-to-Text), audio is sent via the Speech SDK or REST API to the service endpoint. For real-time transcription, audio streams directly to Azure; for batch transcription, audio is uploaded to a storage location you specify, and the service processes it from there.
• Retention Policy:
  By default, Azure does not retain audio or transcripts beyond what is necessary to provide the service. If you disable diagnostic logging (tracing), no audio or transcription data is stored. If tracing is enabled, audio may be temporarily stored in the region you selected, but only accessible to you. Microsoft does not use this data for advertising or unrelated purposes; it is processed solely to deliver the requested service.
• Compliance and Control:
  Customers are responsible for compliance with privacy laws and must configure retention settings. You can also use features like restrictOutboundNetworkAccess and customer-managed encryption keys for additional security.

How Liberty Court Recorder (produced by High Criteria) uses Azure Cognitive Speech Services and whether court audio is retained

• Integration Process:
  Liberty Court Recorder (LCR) works with Post Recording Manager (PRM), which monitors folders where Liberty recordings are saved. PRM’s Speech-to-Text (STT) service sends these recordings to Azure Cognitive Speech Services for transcription. The result is returned as an RTF text file for editing and proofing.
• Audio Handling:
  LCR records multi-channel audio (up to 8 or more channels) to improve speaker separation and transcription accuracy. These files remain on your local or network storage; Azure only processes them temporarily for transcription. Liberty does not store court audio permanently in Azure unless the tick box to store the files form processing is selected in PRM (off by default) — the audio is transmitted for processing and then returned as text. The responsibility for storage and deletion lies wholly with the client local storage network.
• Data Sovereignty:
   Courts typically configure Azure resources in local regions (e.g., Australian data centres) to meet sovereignty requirements. High Criteria recommends setting up an Azure account in-country to meet this compliance.

Key Takeaways

• Azure Cognitive Services processes audio but does not retain it by default unless diagnostic logging is enabled.
• Liberty Court Recorder sends audio to Azure for transcription but keeps original recordings locally; Azure is not used as a temporary staging or a permanent storage solution for court audio.