"VoIP quality is poor" was a legitimate criticism in 2010. In 2026, on a properly configured business connection, VoIP delivers audio noticeably better than a traditional phone line, because HD voice codecs carry roughly twice the frequency range of the old analogue network.
When calls do break up, the cause is almost never the NBN connection speed. It’s one of five things, and all five are fixable. This guide works through them in the order they cause problems.
1. How Much Bandwidth Voice Actually Needs
Far less than most people assume. A single HD voice call consumes roughly 85–100 kbps in each direction including protocol overhead. The commonly used G.711 codec sits near 87 kbps; compressed codecs such as Opus can run considerably lower.
| Concurrent Calls | Upload Needed | Suggested Minimum NBN Plan |
|---|---|---|
| 1 – 4 | ~0.4 Mbps | NBN 50 (20 Mbps up on FTTP/HFC tiers) |
| 5 – 10 | ~1 Mbps | NBN 100 |
| 10 – 25 | ~2.5 Mbps | NBN 100 with QoS, or business fibre |
| 25 – 60 | ~6 Mbps | Business-grade symmetric service |
Note that upload is the constraint, not download. Many Australian small businesses run a residential-style plan with generous download and modest upload, then wonder why callers say "you're cutting out" while their own audio sounds fine. That asymmetry is the classic signature of an upload bottleneck.
2. The Four Metrics That Determine Quality
| Metric | What It Means | Target | Symptom When Bad |
|---|---|---|---|
| Latency | One-way delay to the provider | Under 100 ms; under 150 ms acceptable | People talking over each other, awkward pauses |
| Jitter | Variation in packet arrival timing | Under 20 ms, ideally under 10 ms | Robotic or warbling audio |
| Packet loss | Packets that never arrive | Under 1% | Words or syllables dropping out |
| MOS | Mean Opinion Score, 1–5 composite | 4.0 or above | General "it just sounds bad" |
Latency to an Australian data centre from an Australian business should be low, typically well under 50 ms within the same capital city. If your provider routes calls through Singapore or the United States, you inherit 150–250 ms of unavoidable delay before anything else goes wrong. This is one of the strongest practical arguments for choosing a provider with local infrastructure.
3. Symptom-to-Cause Diagnostic Table
| Symptom | Most Likely Cause | Fix |
|---|---|---|
| Only the far end hears breakup | Upload saturation or Wi-Fi uplink | QoS on upstream; move phones to Ethernet |
| Quality drops at the same time daily | Backups, cloud sync, or CCTV upload | Reschedule backups; rate-limit sync clients |
| Robotic audio on all calls | Jitter, often from Wi-Fi or a cheap switch | Wired connection; enable jitter buffer |
| One-way audio | NAT or firewall traversal / SIP ALG | Disable SIP ALG on the router; check port config |
| Calls drop at ~30 seconds or ~15 minutes | SIP session timer or NAT keepalive mismatch | Adjust registration interval with your provider |
| Echo on some calls | Handset/headset acoustics or a device fault | Test with a different headset; adjust gain |
| Only calls to mobiles sound poor | Transcoding at the carrier interconnect | Raise with the provider; check codec negotiation |
| Static or buzzing on a single desk | Cabling or PoE injector fault | Swap patch lead and port to isolate |
SIP ALG: the single most common misconfiguration
Most consumer and small-business routers ship with "SIP ALG" enabled. It’s intended to help VoIP traverse NAT; in practice it rewrites SIP packets incorrectly and causes one-way audio, dropped registrations, and calls that fail after a fixed interval. Turn it off. Almost every VoIP provider in Australia will tell you the same thing, and it should be the first change you make on any new router.
4. Choosing the Right NBN Service
Three things matter more than the headline speed tier.
- Access technology. FTTP and HFC generally deliver the most consistent latency and jitter. FTTN performance depends heavily on copper distance from the node, a long lead-in can produce enough jitter to affect voice even when speed tests look fine. If you’re on FTTN and eligible for a fibre upgrade, take it.
- CVC/backhaul provisioning. Retail providers buy capacity in bulk; some oversubscribe it. Evening congestion that barely affects streaming is very audible on a call. Business-grade plans generally carry better contention ratios and this is what you’re paying the premium for.
- Business SLA. Enterprise Ethernet and business NBN plans include restoration timeframes. A residential-grade plan may leave you without phones for days after a fault. If your phone system is on the same connection as your revenue, buy the SLA.
A frequently overlooked option: run voice on a separate, small NBN or 4G/5G service from your data. It isolates the phones from whatever the office network is doing and gives you a natural failover path.
5. Setting Up QoS Properly
Quality of Service tells your router that voice packets go first. On a connection shared with file uploads, video calls, and cloud backups, it’s the difference between reliable phones and intermittent complaints.
- Identify your voice traffic, by VLAN if your switches support it, otherwise by device IP address or DSCP marking (voice is normally marked EF / DSCP 46).
- Set an upstream bandwidth reservation, allocate roughly 100 kbps per concurrent call, with headroom.
- Enable strict priority queuing for that class so voice is never queued behind a large upload.
- Cap the total shaping rate to slightly under your real upload speed (around 90–95%). QoS only works when your router, not the carrier, is the bottleneck.
- Create a voice VLAN if you’ve managed switches. It isolates phones from broadcast traffic and makes prioritisation reliable end to end.
- Verify under load: start a large upload and make a test call. If quality holds, QoS is working. If not, your markings aren’t surviving the path.
6. Wi-Fi: The Most Common Culprit
More VoIP quality tickets trace back to Wi-Fi than to the NBN. Wi-Fi introduces variable latency by design, and voice is intolerant of variability.
- Wire the desk phones. Any fixed handset should be on Ethernet with PoE. This isn’t negotiable in a busy office.
- Use 5 GHz or 6 GHz for softphones and avoid the crowded 2.4 GHz band.
- Check for roaming problems. A laptop walking between access points without fast roaming support will drop audio at every handover.
- Enable WMM (Wi-Fi Multimedia) so voice traffic is prioritised over the air, not just on the wire.
- Watch access point density. Too many APs on overlapping channels is as harmful as too few.
7. Failover and Business Continuity
The legitimate concern about cloud phone systems is what happens when the internet drops. A properly designed deployment answers this three ways:
- Mobile app direct routing. When the office connection fails, calls route to staff mobile apps over the mobile network. Nothing is missed and callers see the same business number. This is the single most valuable continuity feature for a small business.
- Automatic overflow to mobile numbers. Configure a failover destination at the platform level so the provider redirects calls if your site deregisters.
- 4G/5G backup on the router. A router with a mobile backup SIM keeps desk phones registered through short outages.
Also confirm how emergency calling behaves. VoIP services must be able to reach Triple Zero, and your registered service address matters for call routing. Make sure your provider has the correct address for each site, and that staff know to state their location, particularly when working remotely.
8. The Pre-Deployment Checklist
- Measure current upload speed at the busiest time of day, not at 7am
- Run a jitter and packet loss test to an Australian endpoint for at least 24 hours
- Confirm the access technology and, on FTTN, the line sync rate
- Disable SIP ALG on the router
- Configure QoS with a voice class and verify under load
- Wire every fixed handset; PoE switch where possible
- Create a voice VLAN if managed switches are available
- Configure platform-level failover to mobile
- Register accurate site addresses for emergency calling
- Run a two-week pilot with a small group before the full cutover
Providers with Australian data centres have a structural advantage on every one of the metrics in section 2, simply because the packets travel less far. When comparing providers, ask directly where your calls are processed. It’s a question about audio quality as much as data sovereignty.