Meeting agents FAQ
Answers to common questions about interactive meeting agents.
FAQ
Why is the bot's audio / video output choppy? How can I improve streaming quality?
If the audio or video output from your bot is choppy, it's likely that your bot's instance doesn't have enough CPU power to handle your use case. You can test this by upgrading the bot to a larger, more powerful instance. Typically the web_4_core instance is sufficient for most Output Media use cases. See Addressing audio and video issues: bot variants for configuration and troubleshooting.
Why doesn't the bot's video/screenshare show in the recording?
Contact us to confirm video/screenshare recording availability for your setup. You can include the bot's audio by setting recording_config.include_bot_in_recording.audio = true. For recording and transcribing the bot's audio, see Recording meeting agents.
Can I use the Automatic Video Output parameter while using Output Media?
Yes. You can use automatic_video_output when your agent uses only output_media.audio. Leave output_media.camera and output_media.screenshare absent. Automatic Video Output cannot be combined with either video target. See Create Bot.
What latency should I expect when using Output Media for speech?
The latency introduced by the a bot's Output Media pipeline should be minimal (in the tens of milliseconds).
That said, total response time includes audio transport, speech recognition (if used), model processing, and speech generation.
Measure how long your agent takes to generate an audio response after receiving meeting audio. For a custom webpage, include the time until the response plays on the page.
If audio is ready at your agent but participants hear it later, contact [email protected] with the bot ID so we can investigate.
How can I make my bot respond only when someone addresses it?
Handle this through your agent or speech-to-speech provider’s conversational controls. Give the model instructions to respond when someone addresses the bot by name or continues a conversation with it, and remain silent during conversations between other participants.
Some providers offer controls specifically for selective responses. For example, Gemini Live’s proactive audio allows the model to decide to remain silent when input is irrelevant. This feature must be enabled on a supported model. Evaluate the behavior with representative multi-participant conversations.
Can I receive participant-specific real-time audio?
Yes. Enable recording_config.audio_separate_raw and configure a WebSocket endpoint subscribed to audio_separate_raw.data. Each audio event includes participant information, allowing your application to process each speaker individually.
The Output Media webpage’s getUserMedia() input provides mixed meeting audio. For participant-specific audio, use the separate audio events delivered to your configured WebSocket endpoint. See Real-time separate Audio per Participant.
How do I build and debug a custom meeting agent?
Start with Build a custom meeting agent and its FAQ. For browser debugging steps, see Debugging Your Webpage.
How do I include the agent's audio in recordings and transcripts?
See recording audio, transcribing the agent's voice, and the recording FAQ.
For more questions, see the custom agent FAQ, webpage FAQ, and recording FAQ.
Updated about 1 hour ago
