Azure provider
Azure providers cover Realtime (Azure OpenAI Realtime deployments), STT (Azure Speech recognition), and TTS (Azure neural voices). Azure is the primary modular STT option for pipeline flows and supports the richest custom endpoint overrides for private or mock Speech services.
Modes
| Flow mode | Provider types |
|---|---|
| Realtime | azure_realtime |
| Pipeline | azure_stt, azure_tts (+ any LLM provider for the middle stage) |
Realtime (azure_realtime)
| Field | Required | Description |
|---|---|---|
| API key | Yes | Azure Speech / AOAI key via team variable |
| URL mode | No | Auto (build from resource name + domain) or Custom (full WebSocket URL) |
| Resource name + domain suffix | Auto mode | e.g. kik-voiceai + cognitiveservices.azure.com |
| Custom URL | Custom mode | Full wss://… base; deployment appended as query |
| Deployment name | Yes (effective) | Azure OpenAI realtime deployment |
| Voice, transcription, VAD | No | Advanced realtime tuning |
Legacy flows may still store a single resourceEndpoint hostname; the editor splits it for display in auto mode.
STT (azure_stt)
| Field | Required | Description |
|---|---|---|
| API key | Yes | Azure Speech key |
| Region | No | Defaults to swedencentral; used when Endpoint URL is empty |
| Language | No | BCP-47 locale (flow language default) |
| Endpoint URL | No | Advanced STT → custom Speech host (DNS or IP) |
When Endpoint URL is set, Delphi connects to that host over WebSocket and applies Azure Speech protocol framing directly. If the path is missing, Delphi adds:
/speech/recognition/{mode}/cognitiveservices/v1
where {mode} is interactive, conversation, or dictation from Recognition mode.
Advanced STT also exposes language detection, partial results, VAD, silence timeouts, and resilience retries.
Example custom STT URLs
wss://swedencentral.stt.speech.microsoft.com/speech/recognition/conversation/cognitiveservices/v1
wss://10.20.30.40:8043/speech/recognition/conversation/cognitiveservices/v1
wss://mock-speech.lab.example:8443/api/v1/speech:recognize
Use the path your gateway documents; Delphi normalises known Azure path patterns to match the selected recognition mode.
TTS (azure_tts)
| Field | Required | Description |
|---|---|---|
| Voice | No | Neural voice id (e.g. en-US-JennyNeural) |
| API key | Yes | Azure Speech key |
| Region | No | Used for default host when Endpoint URL is empty |
| Endpoint URL | No | Advanced Azure TTS |
| URL scheme | Transport |
|---|---|
wss:// / ws:// | WebSocket synthesis (default path /tts/cognitiveservices/websocket/v1 if omitted) |
https:// / http:// | HTTPS REST synthesis |
SSML pass-through is supported when Use SSML is enabled — test audio before publishing.
Pipeline TTS uses the modular speech playback queue. Azure Realtime streams audio from the deployment instead.
Custom endpoints and private speech
Azure STT/TTS are the main providers for mock servers, on-prem Azure Speech containers, and IP-only lab endpoints. See Custom STT/TTS endpoints for proxy, Squid, and troubleshooting guidance.
See also
- Custom endpoints
- OpenAI — common LLM/TTS partner in pipeline mode
- Providers overview