Azure OpenAI Service Migration Challenges for Healthcare Company (Tier 5 OpenAI User)

Question

We're a healthcare company currently on Usage Tier 5 with OpenAI and are planning to migrate our infrastructure to Azure OpenAI Service. We handle sensitive healthcare data and have encountered several roadblocks that need addressing before we can complete our migration:

1. Whisper API Rate Limits: We currently process 3-7K audio transcription requests daily, with peaks of ~300 requests per minute. Our OpenAI tier allows 500 RPM, but Sweden Central deployment only offers 3 RPM by default. How can we increase this limit to match our current usage?
2. Content Filtering for Healthcare Data: Our medical content occasionally gets flagged as "gore" or "inappropriate" (e.g., descriptions of broken bones, severe pain). OpenAI previously adjusted their filters for our healthcare use case. What's the process for requesting similar adjustments in Azure?
3. Nordic Language Support: We've noticed that some models (like o3-mini) don't list Danish, Swedish, or Norwegian as supported languages in Azure, while these same models support these languages on OpenAI. Can you confirm these languages will work as expected in Azure deployments?
4. Model Access Requests: We need access to several models including o1 and o3-mini that currently require special access. What's the fastest way to get these approved for our healthcare workloads?

Has anyone successfully migrated from OpenAI to Azure with similar requirements? Any advice on expediting these requests or alternative approaches would be greatly appreciated!

Answer 1

Hello Peter Toth,

Welcome to the Microsoft Q&A and thank you for posting your questions here.

I understand that you have question on Azure OpenAI Service Migration Challenges for Healthcare Company (Tier 5 OpenAI User).

On Whisper API Rate Limits, you can request quota increase, check regional limits, deploy in high-capacity region, use API Gateway for load balancing. This is how to:

1. Run the following Azure CLI command to check the current quota - availability in Sweden Central: az vm list-usage --location "swedencentral"
2. Submit a Quota Increase Request - https://learn.microsoft.com/en-us/azure/ai-services/openai/quotas-limits
3. Consider deploying in another region, Sweden Central might not support higher rates. Try regions like West Europe or North Europe, which may have higher limits. - https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/regional-availability
4. Implement load balancing for High Traffic by implementing an Azure API Management Gateway in front of the Whisper API and use Autoscaling and Azure Front Door to distribute requests across multiple regions. - https://learn.microsoft.com/en-us/azure/api-management/api-management-autoscale

On content filtering for Healthcare Data - submit Limited Access Request, use Azure Text Analytics, fine-tune a GPT model.

1. Submit an Exception Request for Medical Content - https://learn.microsoft.com/en-us/legal/cognitive-services/openai/limited-access
2. Use Azure Cognitive Services for Content Moderation by deploy Azure Text Analytics to pre-process medical text before sending it to OpenAI. - https://learn.microsoft.com/en-us/azure/cognitive-services/text-analytics/how-tos/text-analytics-for-health
3. Fine-Tune GPT Models to Reduce False Positives by train a custom GPT model using Azure OpenAI Fine-tuning API to improve healthcare-specific responses. - https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/fine-tuning

On Nordic Language Support - though, the o3-mini model in Microsoft Azure OpenAI Service does support Nordic languages. This includes languages such as Danish, Finnish, Icelandic, Norwegian, and Swedish. - https://azure.microsoft.com/en-us/blog/announcing-the-availability-of-the-o3-mini-reasoning-model-in-microsoft-azure-openai-service/ and https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/reasoning. However, you can validate supported languages for o3-mini in Azure, by using this API call to check supported languages:

curl -X POST "https://api.openai.com/v1/models/o3-mini/completions" -H "Authorization: Bearer $AZURE_OPENAI_KEY"

If Nordic languages are NOT Supported, use GPT-4 Turbo Instead because it support more languages and might be a better fit. - https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/models and other options if GPT models in Azure don’t support Nordic languages natively, use Azure Translator before sending data to OpenAI. - https://learn.microsoft.com/en-us/azure/ai-services/translator/

Lastly on Model Access Requests, you can submit request, fast-track approval via Azure Support, and use GPT-4 Turbo if approval delayed. -

I hope this is helpful! Do not hesitate to let me know if you have any other questions or clarifications.

---

Please don't forget to close up the thread here by upvoting and accept it as an answer if it is helpful.