Choppy or dropped VoIP calls: diagnosing audio quality in Belgium

A robotic voice, missing syllables, awkward delay or a dropped call does not prove that “VoIP is bad”. They indicate a real-time stream is encountering a problem somewhere between the microphone and the other party. This guide turns a vague symptom into repeatable tests, timestamped evidence and a useful next action.

Choppy or dropped VoIP calls: diagnosing audio quality in Belgium
Good diagnostics follow the audio path rather than immediately blaming the phone or carrier. GVISION illustration.

The short answer

To diagnose a choppy VoIP call, record the exact time, call direction, both numbers, device and network used. Then reproduce the call over Ethernet and Wi-Fi, internally and externally. Correlate these tests with packet loss, jitter, latency and PBX, SBC, firewall or Teams logs. Evidence for each path is more useful than an isolated speed test.

  • A call can be poor in only one direction: analyse sent and received audio separately.
  • Average bandwidth rarely explains voice quality; consistency, congestion and loss matter more.
  • An Ethernet-versus-Wi-Fi test quickly isolates the local network without changing telephony.
  • QoS helps when a link is busy, but does not fix an unstable line or poor routing.
  • An actionable escalation includes precise times, reference calls and comparison results.

Start with the symptom, not a solution

Describing precisely what each party hears immediately reduces the number of hypotheses. “Poor quality” can mean at least six phenomena that require different lines of investigation.

Choppy voice or missing words

Audio fragments disappear when packets arrive too late, out of order or not at all. If only the remote party hears you poorly, examine the upstream path from your site first. If only you hear them poorly, look at the downstream path. This avoids averaging two streams when only one is degraded.

Robotic or metallic voice

The codec and concealment mechanism try to reconstruct incomplete sound. A brief Wi-Fi disruption, saturated queue or network change can cause this effect without dropping the call. Note whether it happens at the start, during a transfer, when video starts or at a fixed time.

Delay and people talking over each other

A high round-trip delay does not necessarily remove words, but disrupts the natural rhythm. It may come from slow internet access, a VPN diverting media, an unnecessarily long geographical path or busy queues. Ask whether the delay is constant or only occurs at peak times.

Echo, one-way audio or a dropped call

Echo may be acoustic — speaker and microphone — or related to endpoint adaptation. One-way audio often points to the media path, NAT, a firewall or address negotiation. A drop after a consistent duration suggests a timeout, session or signalling issue rather than simple jitter. Record each pattern without jumping to conclusions.

The GVISION page on VoIP telephony for Belgian organisations presents the service and possible architectures. Here, the aim is different: to build a reusable diagnostic process for Teams Phone, a Yeastar PBX, a softphone or an IP phone.

Collect the minimum evidence for each call

A ticket saying “it has been dropping since yesterday” is almost impossible to correlate; a timestamped call with context can be found in logs. Prepare a short form that support can complete in under two minutes.

Keep the local time to the minute and the time zone if several countries are involved, the numbers or accounts concerned, inbound or outbound direction, internal or external call, site, device, headset, application, Ethernet, Wi-Fi, 4G/5G or VPN connection, and who hears the fault. Add a literal description, for example: “the external caller cannot hear me for ten seconds”.

Then ask whether the issue is reproducible and affects one user, a room, a floor, a site, a carrier or all calls. An isolated incident deserves recording, but a pattern helps select the next measurement. Three successful reference calls are as useful as three failures: they show what distinguishes a healthy path.

Do not collect conversation content unnecessarily. Technical metadata is usually enough. If recording is considered, apply internal confidentiality and information rules. The aim is to identify interfaces, not listen to business conversations.

In Microsoft Teams, the Call Quality Dashboard and per-call data help distinguish users, subnets, connection types and streams. Microsoft’s guide toreviewing quality of experience with CQD, updated on 16 June 2026, recommends observing trends and problem areas rather than a single individual complaint.

Latency, jitter, loss and MOS: what the measurements really tell you

No measurement is sufficient alone: interpretation must connect the figure to the stream direction, call time and platform.

Packet loss

Real-time voice cannot wait long for retransmission. Brief loss may be concealed; a burst causes missing syllables or robotic sound. A daily average can hide thirty disastrous seconds. Look for the distribution, peaks and affected direction.

Jitter

Jitter is variation in the delay between packets. The jitter buffer reorders the stream and absorbs reasonable variation at the cost of some delay. If too small, late packets are discarded; if too large, conversation becomes slow. The Yeastar jitter buffer documentation describes this trade-off and distinguishes fixed and adaptive modes.

Latency and round-trip time

Latency affects interactivity. A ping to an arbitrary server does not necessarily represent the actual media path. Measure towards the platform or relevant points and, where possible, separate endpoint-to-network, internet access and service. Microsoft’s connectivity test uses success targets of under 1 % UDP loss, under 100 ms latency and under 30 ms jitter; these thresholds belong to that test, not a universal rule for every carrier or PBX.

MOS

The Mean Opinion Score combines several parameters into an estimated quality score. It helps with triage but may be calculated differently across products. In its QoS report, Yeastar rates a MOS below 3.5 as poor, from 3.5 to below 4 as average and from 4 to 5 as good. Use this classification in context and retain raw measurements to explain the score.

Do not turn a threshold into a verdict

Microsoft classifies audio jitter above 30 ms, among other factors, as a poor-stream indicator in CQD. A call below that threshold can still be poor because of burst loss, the endpoint or carrier path. Conversely, a brief peak does not prove a network redesign is necessary.

Isolate five failure domains

Diagnostics become manageable when each test is assigned to a domain: endpoint, LAN/Wi-Fi, WAN/internet, media security or platform/carrier.

1. Endpoint and audio

Test another headset, USB port, handset and device. Check drivers, CPU load, power saving, firmware and the application. If the fault follows the headset, there is no point changing router QoS. If several devices in the same office fail, work back towards the network.

2. Local network and Wi-Fi

Compare the same call over Ethernet. Check coverage, interference, roaming, channel occupancy, port errors, duplex, PoE power and queues. A strong Wi-Fi signal guarantees neither low contention nor freedom from retransmissions. A wired IP phone and wireless softphone also take different local paths.

3. WAN, internet and VPN

Examine upstream and downstream saturation, loss, route changes, dual connections, SD-WAN and carrier incidents. Cloud backup, synchronisation or a video meeting can fill a queue despite good contracted bandwidth. A VPN that unnecessarily centralises media sometimes adds latency and failure points; check the platform design before making exceptions.

4. Firewall, NAT and SBC

RTP or SRTP media must pass in both directions with the expected ports, addresses and session durations. Inspections, SIP transformations, NAT timeouts or overly narrow ranges can cause one-way audio and disconnections. Microsoft warns that an overly restrictive media port range can cause dropped calls and poor quality. Do not randomly disable a feature such as SIP ALG: first validate the vendor recommendation and topology.

5. PBX, cloud, trunk and telephone network

A healthy internal call and degraded external call point towards the SBC, trunk, carrier or remote termination. Compare inbound and outbound calls, destinations, codecs, transcoding and routes. PBX and provider logs must be correlatable with the reference call. Without numbers and the exact time, analysis often stops at an impression.

Decision matrix: which test should you perform next?

The best test changes one variable and produces a clear interpretation. This matrix avoids simultaneous changes that make results unusable.

VoIP symptoms, comparisons and likely domain
ObservationControlled testWhat the result indicatesLimitationBest next step
Only one userAnother headset, then another endpoint, same network.The fault follows the accessory, endpoint or account.A single failure may be intermittent.Repeat a timestamped reference call.
Wi-Fi poor, Ethernet goodSame device, same remote party, two connections.The wireless LAN becomes the priority.Calls are never perfectly identical.Measure channel, retransmissions, roaming and load.
Internal good, external poorEndpoint-to-endpoint call, then a call to the public network.SBC, trunk, carrier or external destination.Codecs and routes may vary.Compare direction, destination and carrier log.
All calls at a fixed timeCompare with WAN traffic and scheduled tasks.Possible congestion or recurring process.Correlation is not yet a cause.Capture queues, loss and utilisation during the peak.
Audio in only one directionCompare inbound/outbound and media addresses.RTP path, NAT, firewall or negotiation.The endpoint may also select the wrong audio device.Read signalling and counters for both call legs.
Call drops after a consistent durationTime several calls.Likely session, timer or signalling issue.A carrier disconnection may coincide.Compare timers, refresh and SBC/PBX logs.
Best choice according to needStart with the least intrusive test that separates two domains, then document before changing configuration.
Five-step process for diagnosing VoIP call quality
Useful diagnostics progressively narrow the scope before any change. GVISION illustration.

A thirty-minute diagnostic procedure

Thirty minutes will not resolve every incident, but can produce a scope, two clean comparisons and an actionable escalation.

  1. Minutes 0–5: qualify. Record three recent calls, the affected direction, sites, devices, networks and symptoms.
  2. Minutes 5–10: reproduce. Call a stable destination with the same user. Note the exact time and what each person hears.
  3. Minutes 10–15: change one variable. Switch from Wi-Fi to Ethernet or change the headset without changing the account or destination.
  4. Minutes 15–20: compare the path. Test an internal then external call, or an inbound then outbound call.
  5. Minutes 20–25: read the evidence. Open CQD, the PBX/SBC report and the network counters corresponding to the time.
  6. Minutes 25–30: decide. Assign a likely domain, record what has been ruled out and designate the next owner.
Technicians comparing a wired IP phone and a softphone on Wi-Fi
The Ethernet/Wi-Fi comparison must keep the same account, destination and similar timing. GVISION illustration.

A consumer speed test is only a clue. It often measures a nearby server using a few parallel streams, without representing UDP traffic, the route to the service, brief outages or the upstream queue at that precise moment. Add continuous measurement, a reference call and the platform’s tools. Yeastar’s IP Ping tool can, for example, check reachability, latency and loss from the PBX to a relevant target.

When QoS actually improves calls

Quality of service is useful when voice packets compete with other traffic on a link or queue you control. It does not create capacity or guarantee handling beyond your domain.

Start by classifying media correctly, marking packets at consistent points and configuring the equipment that actually queues traffic. Microsoft recommends DSCP 46 for Teams audio and specifies that endpoint marking must be accompanied by corresponding configuration on network equipment. If a switch erases the marking or the router ignores the class, the label adds no value.

Size the queue too. Unlimited strict priority can starve other services; an undersized class drops voice during a peak. On the internet, markings may be changed or ignored. QoS is most decisive in the LAN, on a private WAN, an SD-WAN or a connection where the organisation and carrier have agreed on handling.

Before deployment, identify where congestion occurs. The Microsoft QoS recommendations for Teams explain classification and ranges. Apply them to Teams, not automatically to every PBX. For a SIP trunk or Yeastar, use the vendor’s and carrier’s documented requirements.

GVISION can bring telephony and network and Wi-Fi architecture together so VLANs, queues, coverage, internet access and monitoring align. This step is relevant when tests show a recurring transport issue, not when just one headset is faulty.

Mistakes that prolong VoIP incidents

Changing several settings at once, relying on bandwidth or escalating without a reference call destroys the value of diagnostics.

  • Restarting before taking notes: logs and the useful timestamp disappear.
  • Confusing Wi-Fi signal with quality: a strong radio signal can hide contention and retransmissions.
  • Using just one speed test: it represents neither the media path nor bursts of loss.
  • Applying universal thresholds: tools, codecs and calculation methods differ.
  • Enabling QoS without measured congestion: complexity increases without addressing the cause.
  • Always disabling SIP ALG: behaviour depends on the firewall, PBX and provider.
  • Ignoring the direction of the fault: the two parties may experience different quality.
  • Forgetting successful calls: they form the comparison group.
  • Blaming the carrier too early: no number, time or path is available to investigate.
  • Declaring the issue resolved after one call: test peak periods and the features that were failing.

Frequently asked questions

What internet bandwidth does a VoIP call need?

The codec and platform determine bandwidth, but available capacity does not guarantee quality. You also need to measure loss, jitter, latency and simultaneous congestion.

Is jitter above 30 ms always bad?

Microsoft uses 30 ms in several Teams criteria, but the threshold must remain linked to the tool and scenario. Analyse peaks, loss and the actual experience.

Why does only one party hear dropouts?

Each direction has its own stream. The upstream path may be degraded while the downstream remains healthy. Note who hears the fault and follow both call legs.

Does QoS solve all voice problems?

No. It arbitrates congestion in equipment that honours it. It does not fix a headset, physical loss, poor routing or a carrier outage.

Should SIP ALG be disabled?

Not automatically. Some equipment implements it poorly; other topologies use it. Follow the documentation for the PBX, SBC, firewall and carrier.

Does a good speed test rule out the network?

No. It may hide brief outages, upstream queues, busy Wi-Fi and a different media route. Compare Ethernet/Wi-Fi and use call data.

What information should you send to support?

Exact time, direction, accounts or numbers, site, device, network, symptom, call ID and comparable test results are the useful minimum.

Conclusion: turn a complaint into a verifiable path

Choppy voice becomes manageable when every team observes the same call, at the same time, on the same path. Standardise collection, run two controlled tests, then correlate endpoint, network, security and platform. To scope an analysis or improve multi-site telephony reliability, you can contact the GVISION team in Belgium.

Official sources

Published on . Updated on .

Let’s talk about your project

Want to apply this advice in your organisation? Let's review your needs and next steps together.

Talk to GVISION