[juliantoledo]

Summary

Part 3 of 5 in the Gemini Live voice assistant series (following #65 and #69):

1. PR 1: Refactor tool discovery and execution (refactor(agent): unify AppFunction discovery, input conversion, and execution #65 — merged)
2. PR 2: Live LLM provider layer (LiveLlmProvider abstraction + GeminiLiveProviderImpl) (feat(agent): add LiveLlmProvider abstraction and Gemini Live WebSocket provider #69 — merged)
3. PR 3 (this PR): Full-duplex audio capture/playback (LiveAudioSession + AndroidLiveAudioSession)
4. PR 4: Live session orchestration (LiveAgentOrchestrator)
5. PR 5: Android VoiceInteractionService and Live UI integration

Changes in this PR

• LiveAudioConfig, LiveAudioSession, & LiveAudioSessionFactory: Introduce a provider-agnostic domain-layer abstraction for full-duplex PCM microphone capture (listenToRecording()), software-gated pause/resume (pauseRecording() / resumeRecording()), assistant audio playback (queueAudio(pcmBytes, sampleRateHz)), interruption flushing (stopAudio()), and lifecycle teardown (release()), configured with per-provider sample rates via LiveLlmSession.audioConfig (LiveAudioConfig).
• AndroidLiveAudioSession:
  • Implement LiveAudioSession pairing VOICE_COMMUNICATION recording with USAGE_VOICE_COMMUNICATION playback under MODE_IN_COMMUNICATION and AUDIOFOCUS_GAIN_TRANSIENT so the hardware VoIP DSP (AcousticEchoCanceler and NoiseSuppressor) cancels speaker playback from the microphone input.
  • Propagate @RequiresPermission(Manifest.permission.RECORD_AUDIO) on start(), fail fast with IllegalStateException on invalid lifecycle transitions or denied audio focus, and respect audio focus across recording and playback.
  • Route communication audio to the built-in speakerphone when no external headset/Bluetooth/USB audio device is connected, and restore communication device, audio mode, and audio focus on release().
  • Keep AudioRecord continuously recording across tool calls while gating emissions in software during pauseRecording() or transient audio focus loss.
  • Support dynamic AudioTrack recreation if queueAudio receives a chunk with a different sampleRateHz.
• DataModule: Provide LiveAudioSessionFactory backed by AndroidLiveAudioSession.
• Unit Tests: Add Robolectric unit tests in AndroidLiveAudioSessionTest covering communication mode & audio focus lifecycle, failed audio focus acquisition, invalid start() lifecycle transitions, speakerphone vs. external headset routing, interruption flushing, software pause/resume, dynamic playback sample rates, audio focus loss/gain across recording and playback, and concurrent release() during start().

Test Plan

• ./gradlew spotlessCheck test validateScreenshotTest assembleDebug assembleRelease -PGEMINI_API_KEY=dummy_key