QA: Voice Input — Google ADK

Prerequisites

• Demo is deployed and accessible at /demos/voice
• Railway service showcase-google-adk is healthy (/api/health returns 200)
• OPENAI_API_KEY is set on the Railway service for the dedicated voice runtime at /api/copilotkit-voice (required for the /transcribe endpoint)
• GOOGLE_API_KEY is set on the ADK agent backend for the voice-demo agent (resolves to the neutral _simple_chat agent)
• A modern browser that supports MediaRecorder (Chromium, Firefox, Safari 14+)
• Microphone hardware available (required only for the mic path in section 3)
• A bundled public/demo-audio/sample.wav is present (used for screenshot/preview generation; the in-app sample button no longer fetches it)

Test Steps

1. Basic Functionality

• Navigate to /demos/voice
• Verify the page header "Voice input" is visible
• Verify the sample-audio row (data-testid="voice-sample-audio") is visible
• Verify the caption reads Sample: "What is the weather in Tokyo?"
• Verify the "Play sample" button (data-testid="voice-sample-audio-button") is enabled
• Verify <CopilotChat /> renders a message composer (data-testid="copilot-chat-input")
• Verify the composer shows a microphone button (data-testid="copilot-start-transcribe-button") — this is the authoritative signal that transcriptionService is mounted on /api/copilotkit-voice

2. Sample-audio path (no mic permission required)

• Click the "Play sample" button
• Immediately, the chat textarea (data-testid="copilot-chat-textarea") contains the canned phrase "What is the weather in Tokyo?" (no async round-trip; no "Transcribing…" state)
• The button stays enabled
• Click send (data-testid="copilot-send-button")
• Within 10 seconds, the agent responds with a text answer or a weather-related tool render (WeatherCard, custom-catchall card, or default tool card — depending on which tool-rendering mode is active on this page)

3. Mic path (manual)

• Click the microphone button (data-testid="copilot-start-transcribe-button") in the composer
• Grant microphone permission at the browser prompt
• Speak "Hello" clearly, then click the mic button again (now data-testid="copilot-finish-transcribe-button") to stop recording
• Within 5 seconds, the textarea contains text matching "hello" (case-insensitive)
• Click send
• Agent responds within 10 seconds

4. Error Handling

• Deny microphone permission, click the mic button
• Verify the UI handles permission denial gracefully (no crash, mic button remains visible)

Expected Results

• Sample button click populates the textarea synchronously (no perceptible delay)
• Agent response renders within 10 seconds of send
• No console errors during the successful paths
• Whisper transcription via the mic path returns text resembling what was spoken (deployment must have OPENAI_API_KEY configured)