Thanks — I’m sorry you’re hitting major issues with Gemini Live not hearing or transcribing audio. I’m here to help narrow this down. To troubleshoot quickly, could you provide the following details? • How are you sending audio to the model (microphone streaming, pre-recorded files, browser capture, native client)? • Sample client code or SDK you use to stream audio (browser WebRTC, gRPC, REST, etc.). • Audio format/encoding being sent (sample rate, channels, codec, PCM/float vs Opus). • Whether you see any model-side errors, warnings, or status messages/logs when you stream. • Whether transcription ever works (intermittent) or never works, and whether only transcription fails while other model responses succeed. • A short timestamped example (or approximate timestamps) of a conversation where speech wasn’t recognized, and whether the speech was loud/soft, overlapping, accented, or contained background noise. • The environment: OS, network conditions (high latency/loss?), and whether this occurs on multiple devices/locations. While you gather that, try these quick checks that often resolve transcription issues: • Ensure audio is sent at a supported sample rate (commonly 16 kHz or 48 kHz) and in a supported codec/encoding for the realtime API you’re using. • Confirm audio channel layout (mono vs stereo); many speech models expect mono. • Verify microphone capture volume and that the client actually sends audio packets (monitor outgoing audio bytes/frames). • Check for clipping or silence detection on the client that might drop audio before transmission. • Test with a known-good prerecorded WAV/PCM sample (mono 16-bit, 16 kHz) uploaded/streamed to isolate capture vs model issues. • If you use automatic activity detection, try disabling it briefly to see if that changes behavior. Tell me which client/streaming method and audio format you use and any logs you have; I’ll propose targeted next steps.