Quick answer: a VoIP phone cuts out when voice packets arrive late or never arrive. Compare three numbers with your provider’s own limits: packet loss (RingCentral: no more than 1%; Zoom Phone: 2%), jitter (both: 40 ms) and one-way delay (150 ms in ITU-T G.114). If calls drop at about 32 seconds, suspect SIP ALG or NAT.
This page is for a desk phone, softphone or mobile app on a hosted phone system (RingCentral, Zoom Phone, Microsoft Teams Phone, 8x8) where the call connects but the voice breaks up or the call drops. Thresholds are quoted from each vendor’s help centre or the governing standard (ITU-T G.114, RFC 3261, RFC 4028, RFC 4787, RFC 3550), read on 10 November 2026. If your phone does not ring at all, that is a different fault: start with our guide to why a business phone is not ringing.
Where Zian fits when a call cuts out on a lead
Zian does not repair network audio; the free steps on this page do that. What a cut-out call can cost is the caller, and catching that caller is the job we built Zian AI for. Its AI Receptionist answers inbound calls 24/7, qualifies the caller, books the appointment, transfers to a person or takes a message, and follows up by voice, SMS, email, WhatsApp, Instagram DMs or Facebook Messenger, in 30+ languages. It has set 64,521+ qualified sales appointments. It is not a phone system and does not replace your RingCentral, Zoom, Teams or 8x8 service. Zian is in a partnership-application beta with no published prices, and each of its three listed plans, including Intel + Widgets (Free Forever), starts with a partnership application.
Why is my VoIP phone cutting out? Late packets and lost packets
A VoIP call is a stream of small audio packets; Zoom says “Typically a PPS of around 50 is normal.” Three things go wrong with that stream, and each sounds different.
- Packet loss sounds like missing syllables and words, or audio that drops out entirely for a moment. Microsoft says burst losses “cause audio to cut out completely.”
- Jitter is variation in when packets arrive. 8x8 says it causes “Delays, and distortion, or even ‘packet loss’ if the delay is too long,” and that above 30 ms “Audio degradation becomes likely, such as choppy or robotic audio.”
- Delay (latency) makes people talk over each other. It does not, on its own, make audio cut out. Describing the long delays of satellite links, RingCentral says they “will delay communications but not block them.”
Twilio says its analysis of “hundreds of billions of calls supports the theory that network transport issues contribute most to reports of audio quality degradation for VoIP calls,” which is why the ladder below starts with the network.
VoIP thresholds by provider: what each one publishes
Four hosted phone vendors publish limits for loss, jitter and delay. They do not agree, and 8x8 publishes three different sets across its own help pages. Use your provider’s row. Microsoft’s figures differ in kind: they are the averages at which its Call Quality Dashboard labels a stream “poor”, not a network target.
| Source | Packet loss | Jitter | Delay | Bandwidth per voice call |
|---|---|---|---|---|
| ITU-T G.114 (standard) | Not covered | Not covered | Below 150 ms mouth-to-ear: “essentially transparent interactivity”; 400 ms one-way is the network-planning ceiling | Not covered |
| RingCentral (Quality of Service guidelines) | “should not exceed 1%” | “should not exceed 40ms” | “< 150ms for one-way”, “< 300ms for two-way” | G.711 87 kbps, G.729 31 kbps, G.722 87 kbps, Opus 60–80 kbps (desk phone); Opus 60–80 kbps (app) |
| Zoom Phone (phone statistics tab) | “2% or less is recommended” | “40ms or less is recommended” | “150ms or less is recommended” | “60–100kbps for high-quality voice” |
| 8x8 (Network Best Practices, quick reference) | Target below 1%; ideal zero | Target below 30 ms; ideal below 10 ms | Target below 150 ms one-way; ideal below 100 ms | G.711 90 kbps, G.722 90 kbps, G.729 35 kbps, symmetric |
| 8x8 (Network Best Practices, requirements table) | 0% | Below 20 ms | Below 100 ms to 8x8 data centres; “known to work” up to 150–200 ms if there is no loss | Same as above |
| 8x8 (Effects of Network Conditions; Garbled and Choppy Audio) | Below 0.3%; “Even 3% packet loss can ‘significantly degrade’” a G.711 call | Below 20; above 40 ms “will cause severe voice quality issues” | Below 120 ms; generally not over 150 ms one way | Not stated on these pages |
| Microsoft Teams (CQD “poor stream” averages) | Above 10% | Above 30 ms | Round trip above 500 ms | Audio: 10 kbps minimum, 58 recommended, 76 best performance (each way) |
Read the table as a range, not a pass mark. Among the vendor targets (leaving aside Microsoft’s poor-stream flag), the loosest loss limit is Zoom’s 2% and the strictest is 8x8’s 0%. Measured loss above 1% breaks both RingCentral’s and 8x8’s published targets, though not Zoom’s. Microsoft applies its flags only to streams of more than 500 packets, and adds a caution worth applying to every row: a stream marked poor “does not mean the audio stream was actually of poor quality, nor does it mean the user perceived a quality issue,” because its media stack corrects for a lot of network damage before anyone hears it.
The Cut-Out Ladder: ten causes, one test each
The Cut-Out Ladder is our name for working through the causes in the order that rules out the most for the least effort, starting with one rule.
The Stopwatch Rule: before changing anything, note when the audio fails. Random breaks through a call point at the network (rungs 1 to 4). Audio that dies, or a call that drops, at the same time on every call points at signalling (rungs 5 and 6). One person affected while colleagues are fine points at their device or connection (rungs 3 and 8).
| Rung | Symptom | Likely cause | Test | Fix |
|---|---|---|---|---|
| 1 | Breaks up when someone uploads, backs up or joins a video meeting | Upload saturated; bufferbloat | Ping your provider while a large upload runs; watch delay climb | Reserve upload headroom; prioritise voice (rung 7) |
| 2 | Random gaps on every device, all day | Jitter or loss on the internet path | Run the loss and jitter check below for 2–5 minutes | Take the hop-by-hop result to your ISP |
| 3 | Only people on Wi-Fi cut out | Interference, weak signal, no WMM | Same call on Ethernet, or beside the access point | 5 GHz, WMM on, more access points, current drivers |
| 4 | Cuts out on VPN or behind a corporate proxy; calls drop | Inspection, proxying or TCP fallback | Same call with VPN off, on a home network | Split-tunnel or bypass rules for the provider’s ranges |
| 5 | One-way audio, no audio, or a drop at about 32 seconds | SIP ALG rewriting signalling | Check the router’s SIP ALG setting; test once with it off | Turn SIP ALG off (and recheck after any router reset) |
| 6 | Drops at a fixed later point, such as about 15½ or 29½ minutes | NAT or firewall UDP timeout; failed session refresh | Time several drops; compare with the firewall’s UDP timeout | UDP timeout longer than the phone’s re-registration interval |
| 7 | Fine when the office is quiet, bad at busy times | No QoS, or markings stripped | Capture packets at the edge; look for DSCP 46 | Mark voice EF (46) and honour it on every LAN hop |
| 8 | One desk phone bad, softphone on the same desk fine | PoE budget, Green Ethernet, cable or port | Swap cable and switch port; try another handset | Fix the port or PoE budget; disable Energy-Efficient Ethernet |
| 9 | Worse on international or mobile-to-desk calls | Codec mismatch, transcoding, low-bitrate codec under loss | Read the codec in the call statistics | Avoid G.729 on a lossy link; enable both G.711 laws |
| 10 | Everyone, everywhere, from the same moment | Provider incident | Provider status page | Wait; report it; route urgent calls to mobiles |
Rung 1: is my upload bandwidth too small for my calls?
Each call needs its bandwidth in both directions at once; the per-call figures for each provider are in the table above (RingCentral’s include packet and Ethernet overhead). Bandwidth is not the only upload fault. The one 8x8 names is bufferbloat, “high latency in packet-switched networks caused by excess buffering of packets,” which “can also cause packet delay variation (also known as jitter).” Voice packets wait in the queue behind the upload.
Worked example, using 8x8’s two planning rules. 8x8 says to “assume that at least 35% of your users are on call at any time” and to “have 50% of your available bandwidth free to accommodate any spike in usage.” For an office of 20 staff: 35% of 20 is 7 simultaneous calls. At 90 kbps for G.711, 7 × 90 = 630 kbps of voice. Keeping that within half the link means at least 1,260 kbps (about 1.3 Mbps) of real upload at the busiest hour. Substitute your own headcount and codec.
Test: run the check in rung 2 twice, once with the network quiet and once during a large upload. If round-trip time jumps under upload, the queue is the problem, not the ISP.
Fix: reserve headroom, schedule backups out of hours, and turn on voice prioritisation (rung 7).
Rung 2: how do I check VoIP jitter and packet loss?
8x8’s procedure for audio breaking up right now is two traceroute tests at once, one to google.com and one to the nearest 8x8 media server (its Australian one is 103.239.164.2), for “a minimum of 2 minutes and a maximum of 5 minutes.” If the problem shows on both and sits outside your own network, “contact the ISP and report the latency and/or packet loss issue from the issue hop.” If the loss or delay shows only on the path to 8x8, 8x8’s next step is the fragmentation test in its Network Utility instead.
For a quick reading on a Mac or Linux machine, save 100 pings to your provider (the example uses 8x8’s Australian server) and run this script over the file. It prints loss, average round-trip time and jitter smoothed with the RFC 3550 estimator, which moves one-sixteenth of the way towards each new difference.
ping -c 100 103.239.164.2 > ping.txt
python3 voipcheck.py ping.txt
import re, sys
txt = open(sys.argv[1]).read()
seqs = {int(s): float(t) for s, t in re.findall(r'icmp_seq=(\d+).*?time=([\d.]+)', txt)}
sent = int(re.search(r'(\d+) packets transmitted', txt).group(1))
loss = 100 * (sent - len(seqs)) / sent
rtts = [seqs[k] for k in sorted(seqs)]
j = 0.0
for a, b in zip(rtts, rtts[1:]):
j += (abs(b - a) - j) / 16 # RFC 3550 A.8 smoothing, applied to RTTs
print(f"sent {sent} lost {sent-len(seqs)} loss {loss:.1f}%")
print(f"avg RTT {sum(rtts)/len(rtts):.1f} ms (one-way is roughly half)")
print(f"jitter {j:.1f} ms")
We ran it on a made-up 100-ping file with 3 pings removed and round trips alternating 20 and 80 ms: it reported 3.0% loss, a 49.1 ms average and 57.6 ms jitter. On a live 60-ping run from our server it reported 0.0% loss and 0.0 ms jitter.
Ping measures ICMP round trips, not your phone’s RTP, so treat it as a screening test. The numbers that count are in your phone’s own call statistics: Zoom shows them in the Phone tab during a call, and Teams admins read them per call in the Teams admin center and Call Quality Dashboard.
Rung 3: does Wi-Fi make VoIP calls cut out?
Often. Microsoft says “Wi-Fi deployments don’t typically take into consideration the network requirements for VoIP services and are often a source of poor quality.” Keep wireless drivers current, since many organisations leave them “unpatched for years.” Turn on WMM (Wi-Fi Multimedia), which “provides basic QoS features to wireless networks.” Add access points where signal is weak. Prefer 5 GHz, which has “less background interference and higher speeds,” while remembering it “doesn’t penetrate walls as easily.”
Test: make the same call on a cable or beside the access point. If the cut-outs stop, the radio link is at fault.
Mobile apps add one documented quirk: 8x8 says that on its app in data mode, an incoming call to the phone’s own mobile number “forces the 8x8 Work for Mobile call to drop.” Its fix is Cellular Assist, which puts the call on the cellular network, or using Wi-Fi for app calls.
Rung 4: VPNs, proxies and firewalls that inspect voice
Microsoft lists “Calls keep dropping (might be due to firewall or proxy blockers)” as a reason to optimise a network for Teams, and notes that some VPN setups “prohibit the use of UDP” and force media onto TCP, where late packets make “the audio healer stitch and stretch audio, resulting in audible artifacts.”
RingCentral recommends a proxy bypass for its domains, which it says prevents “the lag, jitter, and dropped calls caused when a proxy tries to inspect real-time voice and video data.” Its Quality of Service guidelines also list functions to disable or bypass for voice, including deep and stateful packet inspection, intrusion prevention, WAN acceleration and UDP flood protection.
Test: the same call with the VPN off. Fix: a split tunnel or bypass rules for your provider’s published address ranges.
Rung 5: why does my VoIP call drop after 30 seconds?
A call that drops at about 30 seconds every time gets that number from the SIP standard. When a phone answers, it sends a 200 OK and waits for an ACK in reply. RFC 3261 says: “If the server retransmits the 2xx response for 64*T1 seconds without receiving an ACK, the dialog is confirmed, but the session SHOULD be terminated.” T1 defaults to 500 ms, so 64 × 0.5 = 32 seconds. Audio can flow normally for those 32 seconds and then the call ends.
The standard does not say why the ACK went missing. Our reading: something between phone and provider is losing or misaddressing it, and on a small-business router the first suspect is SIP ALG. 8x8 explains that SIP ALG makes the router rewrite “the payload section of a SIP Packet to change the Private address to be Public address.” RingCentral warns that enabling it “may cause signaling issues, resulting in compromised call features such as one-way audio or a lack of audio.” RFC 4787, the IETF’s NAT best practice, says that to avoid interfering with NAT traversal and allow integrity protection, “NAT ALGs for UDP-based protocols SHOULD be turned off.”
Test: find the SIP ALG setting (sometimes called SIP transformations or SIP helper) and make one test call with it off. Fix: leave it off. 8x8 adds a warning: “Many routers will re-enable ALG by default if the router is ever reset or powered off then back on.”
Rung 6: calls that drop at 15 minutes or another fixed time
A later drop at the same point each time suggests a periodic signalling message is not getting through. Two standards set the clock.
Session timers. RFC 4028 recommends 1,800 seconds (30 minutes) as the session interval, recommends sending the refresh “once half the session interval has elapsed,” and says that if the refresh “times out or generates a 408 or 481 response, then the UAC sends a BYE.” With the recommended values, that is a refresh at 15 minutes and, if it fails, a hang-up after the 32-second (64 × T1) transaction timeout: about 15½ minutes. If the side waiting for the refresh never sees it, it should send a BYE “slightly before the session expiration,” by the smaller of 32 seconds and a third of the interval: 1,800 − 32 = 1,768 seconds, about 29½ minutes. The arithmetic is ours, using the recommended values; providers may use others.
NAT and firewall UDP timeouts. Routers forget idle UDP mappings. RFC 4787 says the mapping timer “MUST NOT expire in less than two minutes” (shorter timers are allowed only for some well-known ports below 1024) and recommends “five minutes or more.” Vendors tie the timeout to their phones’ re-registration. RingCentral says Cisco phones refresh every four minutes and Poly phones every five, so “you must set the NAT entry expiration timeout to more than five minutes.” 8x8 says its phones “call out to register every 11 minutes, or 660 seconds,” and that the UDP timeout should match. Our reading: when the timeout is shorter, the path back to the phone closes, and the provider’s next message, such as a session refresh, has nowhere to go.
Test: time three dropped calls; if they cluster, compare with your firewall’s UDP timeout. Fix: set the UDP timeout above your phones’ re-registration interval (for SIP over TLS, 8x8 says the TCP timeout instead), and check SIP ALG as in rung 5. RFC 4028 itself warns that “Firewalls and NAT ALGs may be very unforgiving about allowing SIP traffic to pass after the expiration time of the session,” which is why it has the waiting side hang up slightly before expiry.
Rung 7: does QoS stop VoIP cutting out?
On your own network, yes; on the open internet, mostly not. Quality of Service marks voice packets so switches and routers send them first. RingCentral, Microsoft and 8x8 each publish the same value for voice media: DSCP 46, Expedited Forwarding (EF). Microsoft assigns it to Teams audio on source ports 50,000–50,019; 8x8 lists it for Windows with admin rights, Mac, iOS, Android and Poly, but DSCP 56 (CS7) for Windows without admin rights.
8x8 notes “QoS does not apply to the ‘open internet.’” RingCentral says ISPs “may change DSCP markings (for example, from 46 to 0) along the internet path.” What QoS fixes is your own queue: the upload link and Wi-Fi from rungs 1 and 3. Microsoft says QoS helps most with packet loss and “to a lesser degree” jitter and round-trip time, and that “Any part of the path that fails to support your QoS priorities can degrade the quality of calls, video, and screen sharing.”
Test: capture traffic at the network edge; voice should still carry 46. Fix: mark and honour EF on every hop you control.
Rung 8: one desk phone cuts out, the rest are fine
If a softphone on the same desk sounds clean, look at the handset and port. 8x8 requires PoE switches with “sufficient power for all connected IP phones simultaneously.” RingCentral lists Energy-Efficient Ethernet (Green Ethernet), which puts ports into low power after inactivity, as something that “may cause intermittent signaling and media traffic issues.” It also says auto-QoS may cause “distortion or incorrect volume levels” on older Poly phones. 8x8 asks for at least Cat 5, “preferably Cat 6,” cabling.
Test and fix: swap the cable, then the switch port, then the handset; keep whichever swap clears it.
Rung 9: does the codec matter?
It changes how much loss a call can take. 8x8 says a G.711 call can be significantly degraded by 3% loss and that “G.729 codec requires packet loss of far less than 1 percent to avoid audible errors.” Opus, which Zoom Phone uses between Zoom endpoints and RingCentral lists for its app, adapts its bitrate; Zoom says it “will adjust to the available bandwidth on a network, consuming approximately 60-100kbps.” 8x8 recommends enabling both G.711 a-law and μ-law to avoid transcoding, because “Transcoding does increase latency.”
Test: read the codec in the call statistics on a bad call. Fix: on a lossy link, avoid G.729 if you can; on international routes, enable both G.711 laws.
Rung 10: is it the VoIP provider, not me?
If everyone on every network started cutting out at once, check the provider’s status page first: status.ringcentral.com, status.zoom.us (which now redirects to zoomstatus.com) and status.8x8.com all load as status dashboards. If it shows an incident, report it and route urgent calls to mobiles until it clears.
When the ladder is clean and one caller still cuts out, the fault may be on their side. Our guide to robotic or choppy audio on an AI voice agent shows why a carrier’s own monitoring can miss a three-second burst. For how loss and delay turn into a MOS score, see what MOS 4 toll-quality VoIP really takes.
Phone fixed, but still losing callers? Zian’s AI Receptionist answers inbound calls 24/7, books the appointment and follows up by SMS, WhatsApp or email when nobody can pick up. Apply For Partnership
Frequently asked questions
Why does my VoIP call keep breaking up?
Usually packet loss or jitter. Compare them with your provider’s published limits: RingCentral says loss “should not exceed 1%” and jitter “should not exceed 40ms”; Zoom Phone recommends 2% or less and 40 ms or less. Test on a cable before blaming the ISP.
Why does my VoIP call drop after exactly 30 seconds?
Most often, the phone that answered never received the ACK confirming the call. Under RFC 3261, it resends its 200 OK for 64 × T1 (32 seconds at the 500 ms default) and then ends the session. A router’s SIP ALG is the first thing to turn off.
What is an acceptable jitter for VoIP calls?
It depends whose limit you use. RingCentral and Zoom Phone say 40 ms or less. 8x8 gives a 30 ms target in its best-practices quick reference and below 20 ms elsewhere, including that page’s own requirements table. Microsoft’s call quality dashboard flags a Teams audio stream as poor when average jitter exceeds 30 ms.
Is 150 ms of latency too much for VoIP?
No; 150 ms is the edge, not the fault. ITU-T G.114 says that below 150 ms mouth-to-ear, most applications get “essentially transparent interactivity,” and it sets 400 ms one-way as the planning ceiling. High delay causes talk-over, not cutting out.
Should I turn off SIP ALG on my router?
For a hosted phone system, yes. RingCentral warns it may cause one-way audio or a lack of audio, 8x8 publishes steps to disable it for eleven router and firewall lines, and RFC 4787 says NAT ALGs for UDP protocols “SHOULD be turned off.” Recheck after any router reset.
Why does my VoIP call cut out only on Wi-Fi?
Wi-Fi is often not set up for real-time voice. Microsoft recommends 5 GHz where coverage allows, WMM turned on, more access points where signal is weak, and current wireless drivers. If the same call is clean on a cable, the radio link is the cause.
Where every figure on this page comes from
| Figure | Who published it | Link | Date read |
|---|---|---|---|
| Below 150 ms mouth-to-ear “essentially transparent interactivity”; 400 ms one-way planning ceiling | ITU-T (G.114, 05/2003) | itu.int: G.114 | 10 November 2026 |
| Loss not over 1%; jitter not over 40 ms; delay below 150 ms one-way and 300 ms two-way; SIP ALG and inspection functions to disable; EF 46 for VoIP media; ISPs may change 46 to 0; Green Ethernet and auto-QoS warnings | RingCentral | Quality of Service guidelines | 10 November 2026 |
| Per-call bandwidth: G.711 87, G.729 31, G.722 87, Opus 60–80 kbps | RingCentral | Bandwidth and network capacity assessment | 10 November 2026 |
| Cisco re-register every 4 minutes, Poly every 5; NAT timeout over 5 minutes; proxy bypass prevents dropped calls | RingCentral | RingCentral network requirements | 10 November 2026 |
| Zoom Phone: delay 150 ms or less, loss 2% or less, jitter 40 ms or less, about 50 packets per second, 60–100 kbps | Zoom | Accessing meeting and phone statistics | 10 November 2026 |
| Zoom Phone uses adaptive Opus between Zoom endpoints, about 60–100 kbps | Zoom | Calculating Bandwidth Usage for Zoom Meetings and Phone | 10 November 2026 |
| Teams CQD poor stream: jitter above 30 ms, loss above 10%, round trip above 500 ms; Wi-Fi, VPN and TCP guidance; poor does not mean perceived | Microsoft | Use CQD to manage call and meeting quality | 10 November 2026 |
| Teams audio 10 / 58 / 76 kbps; “Calls keep dropping” and “static and cut out” reasons | Microsoft | Prepare your organization’s network for Teams | 10 November 2026 |
| Teams audio DSCP 46 (EF) on ports 50,000–50,019; QoS must be consistent end to end | Microsoft | Implement Quality of Service in Microsoft Teams | 10 November 2026 |
| 8x8 quick-reference targets (1%, 30 ms, 150 ms) and requirements table (0%, 20 ms, 100 ms); 90 / 35 kbps per call; 35% on call and 50% free; 660-second registration; EF 46 and CS7 56; PoE and cabling | 8x8 | 8x8 Network Best Practices | 10 November 2026 |
| Limits below 0.3, 20 and 120 ms; 3% loss degrades G.711; G.729 needs far below 1%; jitter above 30 ms and 40 ms | 8x8 | Effects of Network Conditions on Voice Quality | 10 November 2026 |
| Two traceroutes for 2–5 minutes; Australian media server 103.239.164.2 | 8x8 | Garbled and Choppy Audio Troubleshooting | 10 November 2026 |
| What SIP ALG rewrites; eleven router and firewall lines; re-enabled after reset | 8x8 | Disabling SIP-ALG in Your Router or Firewall | 10 November 2026 |
| Bufferbloat definition: “high latency in packet-switched networks caused by excess buffering of packets”; can also cause jitter | 8x8 | Network Readiness Guidelines and Best Practices | 10 November 2026 |
| Mobile app in data mode drops when the mobile number rings | 8x8 | 8x8 Work for Mobile Calls on Data Mode Drops | 10 November 2026 |
| Network transport issues contribute most to audio-quality reports | Twilio | Voice Insights FAQ | 10 November 2026 |
| T1 500 ms default; 64*T1 without ACK ends the session (32 seconds) | IETF (RFC 3261) | RFC 3261 | 10 November 2026 |
| Session interval 1,800 seconds recommended; refresh at half; BYE on timeout; BYE before expiry by min(32 s, one third) | IETF (RFC 4028) | RFC 4028 | 10 November 2026 |
| UDP mapping not under 2 minutes, 5 or more recommended; NAT ALGs SHOULD be off | IETF (RFC 4787) | RFC 4787 | 10 November 2026 |
| Interarrival jitter estimator, 1/16 smoothing | IETF (RFC 3550) | RFC 3550 | 10 November 2026 |
| Office worked example (7 calls, 630 kbps, 1,260 kbps); 15½ and 29½ minute drop points; script outputs | Zian (our arithmetic and test, from the rows above) | This page | 10 November 2026 |
Ready for the calls your phone system cannot catch? Zian runs AI agents that answer inbound calls 24/7 and follow up across voice, SMS, email, WhatsApp, Instagram DMs and Facebook Messenger. Apply For Partnership