For real-time latency analysis, you can monitor the Iravoice dashboard, which provides several network and streaming metrics, including:

• SSL handshake time

• WebSocket (WSS) connection delays

• WebSocket socket errors

• Other streaming/network-related metrics

A detailed knowledge base article explaining these metrics is already available internally.

https://support.epicode.in/help/en-gb/42-iravoice/287-components-of-iravoice-web-monitor

Post-Call Latency Analysis

After calls are completed, latency and quality metrics can be analyzed using the Call Event CDRs received on the webhook endpoint.

The following events are useful:

1. call_quality

This event provides overall call quality information, including:

• MOS (Mean Opinion Score)

• Quality

This helps evaluate the overall user experience for each call.

2. stream_stopped

This event contains detailed stream-level latency metrics within the "setup_delays" parameter, including:

• Start time

• SSL handshake delays

• Average bot delay

• WSS/WebSocket connection delays

Example: “setup_delays”:  {"started":"2026-05-15T11:39:17.813164Z","resolve":0,"connect":0,"ssl_handshake":4641,"wss_handshake":36,"first_audio":16,"att":0,"awt":0,"abd":13}

Unlike the dashboard, which shows averaged metrics across calls, the stream_stopped event provides per-call visibility, making it useful for debugging specific latency issues.

BotCompose Latency Monitoring

For BotCompose calls, latency analysis can be performed using the BotCompose CDRs.

The CDR contains latency_metrics which shows per-call timing information for:

• LLM response delay

• STT (Speech-to-Text) delay

• TTS (Text-to-Speech) delay

Example:

Additionally, you can review the call_quality event to analyze:

• MOS (Mean Opinion Score)

• Overall call quality metrics

This helps identify whether latency originates from the AI pipeline (LLM/STT/TTS) or from network and media streaming layers.