Why VoIP calls break up: latency, jitter, packet loss and how to fix them
An internet call that sounds robotic, cuts out for a second or lets only one person hear the other is almost never a "slow internet" problem. Voice needs very little bandwidth. It needs packets that arrive quickly, evenly and completely.
The three numbers that matter
Latency
Latency is the time a packet takes from mouth to ear: capturing and encoding the sound, crossing the networks, buffering and decoding. The ITU's planning recommendation G.114 states that most users find one-way delay below about 150 ms acceptable for conversation, and that above roughly 400 ms it becomes unacceptable for general network planning. As delay grows, people start talking over each other, then wait awkwardly for replies.
Jitter
Jitter is the variation in delay from one packet to the next. Packets that leave every 20 ms may arrive after 30, then 80, then 25 ms. The receiving app holds them in a jitter buffer to smooth this out, which adds delay. When jitter exceeds what the buffer can absorb, late packets are discarded and you hear gaps or warbling.
Packet loss
Lost packets mean missing slices of speech. Modern codecs hide occasional losses (packet-loss concealment), and Opus can carry redundant data to recover from them, but bursts of loss produce the typical "robot voice" or chopped words.
Match the symptom to the cause
| What you hear | Likely cause | First thing to try |
|---|---|---|
| Robotic or metallic voice, words chopped | Packet loss or high jitter | Move closer to the router, or switch from Wi-Fi to mobile data to compare |
| Long pauses, talking over each other | High latency, often a relayed or congested path | Stop large uploads/downloads; avoid VPNs routed far away |
| You hear them, they cannot hear you (one-way audio) | NAT/firewall problem, often a router "SIP ALG" | Disable SIP ALG on the router; check app microphone permission |
| Call drops after exactly the same time (e.g. 30 s or 15 min) | Router or firewall closing the UDP session | Update router firmware; disable SIP ALG; check provider's keep-alive advice |
| Echo of your own voice | The other person's speaker feeding their microphone | Ask them to use earphones or lower speaker volume |
| Quality drops when someone starts streaming or a backup runs | Upload congestion ("bufferbloat") | Enable smart queue management on the router |
Wi-Fi: the usual weak link
- Distance and walls. Every wall, floor and mirror costs signal. A weak link forces retransmissions, which add delay and jitter even when a speed test looks fine.
- Band choice. 2.4 GHz travels further but is crowded and shared with Bluetooth, microwaves and neighbours' networks. 5 GHz (and 6 GHz on Wi-Fi 6E/7) is less congested at shorter range. If your phone sits on a weak 2.4 GHz connection, moving into a room with the router can transform call quality.
- Roaming between access points. Mesh systems and multiple access points can briefly interrupt a call as you walk between them. Standards such as 802.11r (fast transition) reduce the gap where both router and phone support it.
- Prioritisation. Wi-Fi has a built-in priority scheme, Wi-Fi Multimedia (WMM), with a dedicated "voice" access category. Leaving WMM enabled on the router, as it is by default on most equipment, helps voice packets go first.
Router settings worth checking
SIP ALG
Many home routers include a SIP Application Layer Gateway that rewrites SIP messages as they pass through, supposedly to help with NAT. In practice, numerous VoIP providers advise switching it off because broken implementations cause one-way audio, failed registrations and calls that drop after a fixed time. It only affects SIP-based services (desk phones, softphones, some providers); app calls with their own protocols are generally unaffected. The setting is usually in the router's advanced, NAT or firewall menu.
Upload congestion and bufferbloat
When your upload link is saturated, by a cloud backup or a large file being sent, router queues fill and every packet, including voice, waits behind them. Smart queue management algorithms such as FQ-CoDel (RFC 8290) and CAKE keep queues short and let small, time-sensitive packets through. Some routers label this "SQM", "Smart QoS" or "adaptive QoS".
DSCP marking
VoIP equipment can mark voice packets with the "Expedited Forwarding" class (DSCP 46, defined in RFC 3246; see the configuration guidance in RFC 4594). Your own network can honour these marks; the public internet usually ignores them. On a business network with a managed router, prioritising marked voice traffic is worthwhile.
Mobile data
On 4G and 5G, app calls compete with other traffic in the cell. Quality usually suffers at the edge of coverage, inside buildings, and in crowded places such as stations and stadiums. If your phone supports it, carrier VoLTE is managed by the operator with its own priority, which is why a regular call may succeed where an app call struggles.
VPNs
A VPN sends every packet through a server first. If that server is far away, latency rises; if it is overloaded, jitter rises. Some VPNs also handle UDP poorly. When diagnosing a bad call, try once with the VPN off.
A quick diagnostic routine
- Make the same call on Wi-Fi and on mobile data. If only one is bad, you know which network to fix.
- Repeat next to the router. If quality returns, the problem is Wi-Fi coverage, not your internet connection.
- Pause uploads, backups and streaming in the household and try again.
- For SIP services, disable SIP ALG and reboot the router.
- Run a test that reports latency and jitter under load, not just download speed.
- If everything else is clean, test with another contact: the problem may be on their side.
For the basics of how voice packets travel, see what is VoIP; for headsets and phone settings, see device support for calling apps.