Voice Live API with gpt-realtime -datazone models: HDS certification scope, EU-only processing and data retention

We run a health-related application hosted in Azure France Central under an HDS (Hébergeur de Données de Santé) contract. We are evaluating the Voice Live API (Azure Speech in Foundry Tools) for real-time voice conversations in French, using a data zone model: gpt-realtime-1.5-datazone on a France Central resource, or gpt-realtime-2.1-datazone on a Sweden Central resource. The audio stream may contain personal health data, so we need to clarify the following before going further.

1. HDS scope Are the Voice Live API, Azure AI Speech and Microsoft Foundry (including the Azure OpenAI realtime models) covered by Microsoft's HDS v2.0 certificate for France Central and Sweden Central? Where can we find the list of in-scope services?

2. Processing location With a -datazone model, are audio, prompts and responses processed exclusively inside the EU, including under load balancing? What is the processing scope of azure-speech input transcription and of HD voices (e.g. fr-fr-Vivienne:DragonHDLatestNeural) for a France Central resource? The Voice Live regions table does not state it for azure-speech.

3. Data retention and abuse monitoring What is stored during a Voice Live session with a -datazone model (audio, transcripts, prompts, responses, function calls), where, and for how long? Does Azure OpenAI abuse monitoring, including possible human review, apply to models served through Voice Live? If our subscription is approved for modified abuse monitoring through the Limited Access Review form (ContentLogging=false), does that also apply to Voice Live pre-deployed models, or only to our own Azure OpenAI deployments (BYOM)?

4. Content filtering Since Voice Live content filtering cannot be modified, is there a supported way, other than BYOM (preview), to avoid blocking legitimate medical descriptions such as seizures, medications or symptoms?

Pointers to the relevant documentation or contractual references for each point would be much appreciated. Thank you.