The fact that both inbound and outbound voice is affected should make this easier to troubleshoot. Whether the problem is with your network, at Amazon or with your trunking provider, you should see the jitter and/or packet loss on traffic leaving the PBX. Based on your “recordings are good” observation, I’d suspect severe jitter.
Start by capturing all inbound and outbound traffic; see
Ask your agents to record the time when trouble occurs and to report it promptly. Look at the timestamps on the capture files, locate the one with the bad calls, download it to your PC and analyze it with Wireshark. If RTP forwarded from extension to trunk is bad, but not vice-versa, the problem is likely with your network or ISP. If the opposite, I’d suspect the trunking provider. If both are bad, either your instance isn’t being scheduled timely, or some resource in your instance is being exhausted at the time.
If the trouble is at the PBX, try running this script:
to see whether the instance is getting serviced when needed.