Deploy Azure AI Speech with CognitiveResource or Microsoft Foundry

Azure Speech today is exposed in two closely related ways that share the same underlying resource provider (Microsoft.CognitiveServices):

1. As a classic Speech resource (kind Speech)
2. As Azure Speech in Foundry Tools (kind AIServices under the same provider)

From the provided information:

• Azure Speech in Foundry Tools is positioned as the recommended entry point for many scenarios because it unifies Speech with other AI capabilities (OpenAI, Vision, Language, etc.) under the Foundry experience and management model.
• Foundry resources share the Microsoft.CognitiveServices provider namespace with Azure Speech, Azure OpenAI, Azure Vision, and Azure Language, and therefore share management APIs, RBAC patterns, networking, and policy behavior.
• Foundry adds extra capabilities around agents, evaluations, and integrated tooling, and the Speech documentation explicitly highlights using Speech “in Foundry Tools” for scenarios like speech-to-speech with Azure OpenAI.

What this means for new deployments:

• For new greenfield solutions, using a Foundry resource (kind AIServices) as the starting point is recommended, then enabling “Azure Speech in Foundry Tools” within that resource. This gives:
  • Integrated management with other AI services
  • Built-in support for Foundry features (agents, evaluations, etc.)
  • Simplified authentication and deployment flows in the Foundry portal
• Existing Speech resources created as Microsoft.CognitiveServices/account with kind Speech remain valid and are still used in documentation and samples (for example, custom speech deployment via REST, CLI, and the Foundry portal can attach to an existing Speech resource).

On deprecation:

• The context explicitly states that Foundry resources and Speech resources share the same provider and management patterns, but it does not state that Microsoft.CognitiveServices/accounts (classic Speech resources) are deprecated or scheduled for deprecation.
• Instead, it positions Foundry as the recommended starting point “for many scenarios,” not as a mandatory replacement.

Practical guidance:

• For new projects: prefer creating a Foundry resource (Microsoft.CognitiveServices/account with kind AIServices) and using Azure Speech in Foundry Tools for speech-to-text, text-to-speech, and custom speech workflows.
• For existing Speech workloads: continue using the current Speech resource; it can be connected into the Foundry portal for fine-tuning and deployment management where needed.

References:

• Microsoft Foundry architecture
• What is Azure Speech?
• Azure Speech features in Microsoft Foundry (new)
• Deploy a custom speech model (ai-foundry-portal)
• Deploy a custom speech model (speech-cli)
• Deploy a custom speech model (rest-api)
• Azure OpenAI speech to speech chat (programming-language-csharp)
• Azure OpenAI speech to speech chat (programming-language-python)