Latency Optimization

Minimize delay between user speech and agent response.

Understanding Latency#

Total latency = STT + LLM + TTS + Network

The LLM is usually the largest share, followed by STT and TTS. Call History shows the per-component breakdown for each call.

Optimization Strategies#

1. Use Native Audio Models#

Bypass STT/TTS entirely:

{
  "llmProvider": "gemini-live-2.5"
}

Native audio models (Gemini Live, OpenAI Realtime) skip separate STT and TTS hops, so they usually respond faster than the traditional pipeline.

2. Choose Fast Providers#

Fastest STT:

• Deepgram Nova-2

Fastest TTS:

• Cartesia
• Deepgram
• Sarvam

Fastest LLM:

• Gemini 2.5 Flash-Lite

3. Optimize LLM Response#

{
  "llmConfig": {
    "maxTokens": 150,
    "temperature": 0.5
  }
}

Shorter responses = faster TTS.

4. Reduce Network Latency#

• Use nearest region
• Check for network issues
• Consider dedicated connection

Measuring Latency#

Check latency in Call History:

1. Open a call
2. View Latency Breakdown
3. See per-component timing

Quick Wins#

1. Switch to Gemini Live
2. Use Deepgram STT
3. Keep responses short