View a markdown version of this page

Troubleshoot audio quality issues in Connect Customer - Amazon Connect Customer
Before you begin: gather this informationStep 1: Identify the symptomStep 2: Determine the scopeStep 3: Localize the problem using the call pathStep 4: Collect diagnostic data (agent-side, paths 1, 2, and 3)Step 5: Resolve agent-side issuesStep 6: Resolve end-customer/telephony issues (paths 4/5/6)Step 7: Before you contact AWS Support

Troubleshoot audio quality issues in Connect Customer

Audio quality issues—choppy or robotic audio, echo, delay, one-way audio, humming, or dead air—can originate anywhere along the call path. This page is the starting point for diagnosing them. It helps you narrow down where in the path the problem occurs, then directs you to the specialized topic that resolves it.

Call disconnects

This page covers audio quality. If calls are dropping or disconnecting rather than sounding bad, see Troubleshoot call disconnects by using DisconnectDetails in the contact record.

Intended audience

This guide assumes you are an IT administrator with experience investigating network, telephony, and workstation issues, and that you can access data in a Connect Customer contact record.

Before you begin: gather this information

Collect the following for the affected calls before you start. You need it to investigate and, if necessary, to open a case with AWS Support.

  • Description of the symptom—what the audio sounds like (choppy, robotic, echo, one-way, humming, dead air, delay) and whose voice is affected. You use this in Step 1.

  • Instance Amazon Resource Name (ARN)—see Find your Connect Customer instance ID or ARN.

  • ContactId of each affected call (provide 3–5 examples, none older than 24 hours).

  • Time of occurrence, including time zone (record in Coordinated Universal Time (UTC) where possible).

  • Call recording for each example.

If you can't obtain a recording directly, ask the agent or customer to review one and confirm:

  • Whose audio is degraded—the agent or the end-customer.

  • Whether the reported degradation is actually audible in the recording.

Prerequisites

Before troubleshooting, make sure of the basics: the agent workstation meets the minimum hardware requirements and the agent uses a supported browser.

Step 1: Identify the symptom

Use the symptom to jump to the most likely cause. If your symptom isn't listed, or you're not sure, continue to Step 2.

Symptom Most likely area Where to go
No audio ("dead air") on the first call after a Windows 11 reboot Workstation (network interface card/services) CCP audio issues on first call after system reboot when using Windows 11
No audio in both directions (neither party can hear the other), not after a reboot Contact Control Panel (CCP) setup, microphone access, or ports Contact Control Panel (CCP) Issues, How do I troubleshoot stopped audio during calls in my Amazon Connect CCP? (AWS re:Post)
One party can't hear the other (one-way audio) Exclusive mic/speaker control, or network address translation (NAT) One-way audio from customers, Troubleshoot your network for call quality and disconnect problems
Persistent humming in the agent's audio Headset/browser sample-rate mismatch Humming sound in the agent's audio device: Verify the headset and browser sample rates
Missed calls / "Microphone is not accessible" / "Initialization Failed" / Web Real-Time Communication (WebRTC) timeout CCP setup, permissions, or ports Contact Control Panel (CCP) Issues
Echo (agent hears their own voice) Mic/speaker feedback Troubleshoot an agent's workstation for call quality and disconnect problems
Choppy/broken, delayed, or distorted/robotic audio Network or workstation Continue to Step 2

Step 2: Determine the scope

The number of affected agents changes where you look first.

  • A single agent—focus on that agent's workstation (Step 5, workstation path) and headset.

  • Multiple agents in the same location or hierarchy—suspect a local network issue (router, internet service provider (ISP), or local area network (LAN)) or a software or OS update pushed to those machines.

  • Agents across multiple locations (remote and in-office)—suspect an organization-level network change or a browser/OS auto-update.

Use AgentHierarchyGroups and DeviceInfo from the contact record to identify the impacted population. For more information about impact analysis, see Troubleshoot audio quality issues by using QualityMetrics in the contact record.

Tip

If the issue started on a specific date for all affected agents, check for network infrastructure changes, browser auto-updates, or OS patches pushed on that date.

Step 3: Localize the problem using the call path

When agents use the Contact Control Panel (CCP), calls traverse this path:

(1) Headset → (2) Agent device/CCP → (3) Agent network → (4) Connect Customer → (5) Telephony network → (6) End-customer device

Connect Customer records each call at point (4), so the recording captures the audio exactly as it arrived at Connect Customer. The recording is a dividing line. If the degradation is audible in the recording, the audio had already degraded before it reached Connect Customer (paths 1–3). If the recording sounds clean but a participant reported bad audio, the degradation occurred after point 4, on the path to the listener.

Use the following table to pair what you observe with the conclusion and where to go next.

Whose audio is degraded Degradation is in the recording Degradation is not in the recording
Agent (the agent's voice, as heard by the customer) The audio degraded before reaching Connect Customer—headset, device, or agent network (1, 2, and 3). Go to Step 4. Audio reached Connect Customer cleanly but customer heard degradation — telephony network or end-customer side (5 and 6). Go to Step 6.
End-customer (the customer's voice, as heard by the agent) The audio degraded before reaching Connect Customer on the inbound side — telephony network or end-customer device (5 and 6). Go to Step 6. Audio reached Connect Customer cleanly but agent heard degradation—agent network, device, or headset (1, 2, and 3). Go to Step 4.
Multi-party and conference calls

For conference or multi-party calls, apply this recording-localization logic per leg (per ContactId), not to the call as a whole.

No recording available

If you can't determine the affected path from the recording—or no recording is available—continue to Step 4 to pull the metrics, which can localize the issue independently.

Step 4: Collect diagnostic data (agent-side, paths 1, 2, and 3)

Run these in order. Each links to its full procedure.

  1. Endpoint Test Utility—from the affected agent's machine, validate WebRTC support, media-device access, per-Region latency (target under 300 ms), and required ports. Download the JSON results. See Validate connectivity to Connect Customer with the Endpoint Test Utility.

  2. QualityMetrics from the contact record—call DescribeContact and review:

  3. CCP logs—if available, download them and open them in the CCP Log Parser. Check for ERROR-level entries and WebRTC metrics during the affected call: PacketLoss, JitterBufferMillis (> 30 ms is abnormal), RoundTripTime (> 300 ms is abnormal). See Download and review Connect Customer Contact Control Panel (CCP) logs.

  4. Amazon CloudWatch—check whether ToInstancePacketLossRate exceeded 20% during the time of the issue. For more information about checking this metric, see Troubleshoot Amazon Connect audio quality issues (AWS re:Post). Audio quality can degrade at packet loss rates well below 20%; this threshold indicates a severe issue requiring immediate network team engagement.

Then branch on what you found:

  • Abnormal metrics found → treat as a network issue (Step 5, network path).

  • No abnormal metrics → treat as a workstation issue (Step 5, workstation path).

Step 5: Resolve agent-side issues

Network (path 3)

Abnormal PacketLoss, JitterBuffer, RoundTripTime, or a high ToInstancePacketLossRate point to the agent network. Check:

  • Virtual private network (VPN)—does the issue reproduce without VPN (direct connection)? If VPN is required, is split-tunneling enabled for real-time traffic?

  • Wi-Fi vs. wired—does it reproduce on a wired connection?

  • Firewall/proxy/NAT—is UDP 3478 (media), TCP 443, and websocket traffic allowed? Use static NAT with keep-alives where possible.

  • Bandwidth contention—are large file transfers or bandwidth-heavy apps running concurrently?

  • Distance to Region—is the agent far from the AWS Region of the instance? (Correlates with high RoundTripTime.)

Full procedure: Troubleshoot your network for call quality and disconnect problems. For more information about metric interpretation, see Troubleshoot audio quality issues by using QualityMetrics in the contact record.

Workstation (paths 1/2)

No abnormal network metrics points to the headset, device, or software. Check:

Full procedure: Troubleshoot an agent's workstation for call quality and disconnect problems and Improve call quality on agent workstations in Amazon Connect contact centers (Prescriptive Guidance).

Step 6: Resolve end-customer/telephony issues (paths 4/5/6)

When the degradation is on the end-customer/telephony side, you can't fix it on the agent workstation. Try to isolate the source before opening a case:

  • Change the calling environment—have the end-customer try a different device, network, or carrier.

  • Check for a common factor—does the issue correlate with a specific carrier, direct inward dial (DID) number, or geographic region? If it correlates with a particular country or number type, confirm that country's coverage and any known telephony restrictions in the Amazon Connect Telecoms Coverage guide on the Amazon Connect website.

  • Check for call forwarding—is another system forwarding calls to Connect Customer? If so, does the issue occur on direct dials to Connect Customer without forwarding?

  • Test multiple numbers—try multiple destination and source numbers to see whether the issue follows a particular number.

If the issue persists across these changes, open a case with AWS Support.

Step 7: Before you contact AWS Support

If the issue persists after troubleshooting, open a case with 3–5 examples no older than 24 hours, and include:

  • Instance ARN and a description of the symptom (whose voice, and how—choppy, no audio, echo, and so on).

  • ContactIds, timestamps (UTC), and contact-record snapshots.

  • Call recordings attached to the case.

  • For end-customer issues: the end-customer phone number (last 4 digits might be masked) and the Connect Customer phone number.

  • Test results: browsers tested, networks tested, alternate-machine test, Endpoint Test Utility JSON export, and your observations after running Ping and MTR.

  • Agent environment details: VPN/firewall/VDI configuration, headset type, and Audio Enhancement mode.

  • CCP type (default vs. custom) and the downloaded CCP logs for the affected calls.

  • Frequency of the issue and the date/time it started (UTC).

Download CCP logs immediately after affected calls

CCP logs persist only within the current browser session. If the agent closes or refreshes the CCP tab, the browser discards the logs—download them as soon as possible after an affected call.