Support Azure AI Foundry models for custom STT and TTS providers
C
CustomWeb AI
Please extend Vapi’s Azure integration beyond LLM models so customers can use Azure AI Foundry / Azure-hosted models for the complete voice pipeline:
- Speech-to-text
- Text-to-speech
- LLMs
Customers should be able to select an Azure AI Foundry deployment for STT or TTS even when that specific model is not yet available as a native Vapi provider.
For example:
An organization deploys or accesses GPT-Live-Transcribe through Azure AI Foundry, then connects that deployment to Vapi as the assistant’s realtime transcriber. Vapi would continue handling the phone call, turn-taking, tools, and orchestration, while Azure handles transcription.
The integration should support:
- Azure AI Foundry deployment names and regional endpoints
- Azure authentication through managed credentials or API keys
- Streaming audio input for realtime STT
- Partial and final transcript events
- Streaming TTS audio output
- Configurable audio codecs, sample rates, and formats
- Language and pronunciation settings
- Custom model/deployment selection
- Regional routing and data-residency controls
- Clear timeout, retry, and provider-error reporting
This would let customers use models that Vapi does not yet offer natively while keeping Vapi’s call handling and orchestration. It would also make Azure-hosted voice models practical for regulated deployments that require EU or Swiss-region processing.
Please clarify whether this should be implemented through:
- A first-class Azure AI Foundry STT/TTS provider,
- A general custom streaming STT/TTS endpoint,
- Or both.