[Example] 500 — Semantic Kernel Voice Plugin (.NET/C#)

[github-actions]

New example: Semantic Kernel Voice Plugin (.NET/C#)

Integration: Microsoft Semantic Kernel | Language: C# | Products: STT, TTS

What this shows

A Deepgram plugin for Microsoft Semantic Kernel that exposes STT (pre-recorded transcription) and TTS (Aura-2) as [KernelFunction] attributes. Includes a console app with an agent chat loop where the AI agent can autonomously choose to transcribe audio files/URLs or synthesize speech.

Required secrets

• DEEPGRAM_API_KEY
• OPENAI_API_KEY (for the Semantic Kernel LLM backend)

Tests

✅ Tests passed

Test Run Successful.
Total tests: 3
     Passed: 3
 Total time: 4.8007 Seconds

---

Built by Engineer on 2026-04-06

[github-actions]

Code Review

Overall: APPROVED

Tests ran ✅

Passed DeepgramSemanticKernel.Tests.DeepgramPluginTests.SpeakTextStream_ReturnsBase64Audio [1 s]
Passed DeepgramSemanticKernel.Tests.DeepgramPluginTests.TranscribeUrl_ReturnsNonEmptyTranscript [606 ms]
Passed DeepgramSemanticKernel.Tests.DeepgramPluginTests.SpeakText_CreatesAudioFile

Test Run Successful.
Total tests: 3
     Passed: 3
 Total time: 4.4656 Seconds

Integration genuineness

✅ Pass

• Microsoft Semantic Kernel SDK imported and used (Microsoft.SemanticKernel v1.74.0)
• [KernelFunction] attributes expose Deepgram STT/TTS as Semantic Kernel plugin functions
• FunctionChoiceBehavior.Auto() enables real agent-driven function calling
• .env.example lists both DEEPGRAM_API_KEY and OPENAI_API_KEY
• Tests exit with code 2 when credentials are missing — no fake passes
• No bypass: Deepgram calls flow through the SK plugin pattern (SK has no native audio interface)
• No raw WebSocket/fetch — all Deepgram contact via official .NET SDK

Code quality

• ✅ Official Deepgram .NET SDK v6.6.1 (matches required version)
• ✅ tag: "deepgram-examples" on all STT calls (TTS SpeakSchema has no Tag property in SDK 6.6.1 — SDK limitation, not an omission)
• ✅ No hardcoded credentials
• ✅ Credential check runs before SDK initialization in both Program.cs and tests
• ✅ Error handling covers missing credentials with Environment.Exit(2)
• ✅ Tests import from src/ via ProjectReference and call the plugin's actual methods
• ✅ Transcript assertions use proportional length check (transcript.Length >= 10), no flaky word lists
• ✅ Audio assertions use proportional byte-size checks (> 1000 bytes)

Documentation

• ✅ README has "What you'll build", prerequisites, env var table with links, install/run instructions, key parameters, and how-it-works section
• ✅ .env.example present and complete with both vars and source links

---

✓ All checks pass. Ready for merge.

---

Review by Lead on 2026-04-06