Troubleshooting the voice agent tester

The troubleshooting tips can help you diagnose problems in your voice agent tester environment.

• When you write test cases, it's easier to use the substring type instead of matching full text with the string type due to inaccuracies with speech to text services.
• When you run a single job from a worker, the job might not fail. However, when you run a batch job for the same worker with multiple jobs, the jobs fail. In this situation, ensure that the Voice Gateway environment variable of MEDIA_RELAY_LOG_LEVEL isn't set to TRACE. If the MEDIA_RELAY_LOG_LEVEL environment variable isn't set to TRACE, you might have more calls that are running concurrently than the caller Voice Gateway can handle. To solve the problem, scale up your deployments with a container management tool like Kubernetes.
• The caller might think that the person on the other end of the conversation stopped talking even though the voice agent tester is pausing between sentences or for some other reason. This situation can occur, depending on the amount of time that the voice agent you're testing takes between turns. You might need to adjust the Voice Gateway POST_RESPONSE_TIMEOUT environment variable either in the caller Voice Gateway or on the worker that starts the calls.
• If long pauses exist to make the voice agent sound more human, make the amount of time before it's the caller's turn to talk longer. Adjust the Voice Gateway WATSON_STT_FIRMUP_SILENCE_TIME environment variable to increase the amount of time.