If a voice bot does not respond or frequently times out, the issue is typically related to delays in the Automatic Speech Recognition (ASR) layer.

ASR converts voice input into text. Delays at this stage can prevent downstream processing, resulting in silence or timeout behavior.

How It Works

Voice Input → ASR → Bot Processing → Response → TTS

Issues usually occur when:

• ASR response is delayed
• Backend times out before ASR returns a result

Common Causes

• High ASR latency

  • Normal: < 4s
  • Degraded: > 5–6s
  • Critical: > 8s

• Traffic spikes

  • Sudden increase in load can cause queuing and delays

• Timeout misconfiguration

  • Backend timeout is lower than ASR response time

• Input-specific delays

  • Large, noisy, or unsupported audio inputs
  • Language or model mismatches

Resolution Steps

1. Check ASR Latency

• Review logs
• Monitor Avg, P95, P99 latency
• If latency > 5–6s → investigate scaling

2. Identify Impact

• Single partner → configuration or traffic issue
• Multiple partners → platform issue

3. Review Traffic

• Compare current vs baseline
• Identify spikes or bursts
• Scale resources if needed

4. Update Timeout Settings

• Ensure timeout > ASR latency
• Recommended buffer: 8–10 seconds under load

5. Validate Inputs (If Isolated)

• Check audio format and size
• Verify language/model configuration
• Review failed requests

Escalate If

• Latency is high without traffic increase
• Failures occur for specific inputs only
• Scaling does not resolve the issue

Provide:

• ASR endpoint
• Sample request IDs
• Time window

Key Point

• Do not treat this as a bot issue until ASR latency is validated.

• Most failures are caused by latency or timeout mismatches, not bot logic.