Does VoIP Work Over Starlink? Latency, Jitter and Setup

Yes. VoIP works over Starlink: Starlink publishes a 25 to 60 ms latency range for its US plans, and SpaceX reported a 25.7 ms median peak-hour latency for US customers in 2025, far inside the 150 ms one-way budget in ITU-T G.114. The catch is consistency rather than speed. Satellite handovers every 15 seconds, a narrow upload lane and queue buildup on the Starlink router produce choppy calls, and each has a fix.

This article covers only the satellite-specific questions. For bandwidth planning and the rollout checklist, see our guide to business VoIP requirements.

Why Starlink latency is so much lower than older satellite internet

Geostationary satellites (HughesNet, Viasat) sit at 35,786 km above the equator. Propagation alone takes 238 to 284 ms for the trip up to the satellite and back down to the ground, so about half a second for a full round trip, and with encoding and ground-segment delays an IP round trip of around 660 ms is common. That exceeds the 400 ms one-way ceiling ITU-T G.114 sets for network planning.

Starlink satellites orbit at roughly 550 km (newer V2 shells sit at 525 to 535 km). At that altitude the up-and-back trip is about 3.7 ms, and the minimum propagation round trip through a satellite to a ground station works out to roughly 7.4 to 15 ms depending on elevation. Add scheduling, the ground network and the hop to the nearest point of presence, and observed minimums land around 20 ms, with a global median of 40 to 50 ms in independent measurements. SpaceX’s 2025 network update reported a 25.7 ms US peak-hour median.

What real Starlink latency, jitter and packet loss numbers look like

Latency. Starlink’s US consumer information sheets list 25 to 60 ms for Residential and Priority plans and “under 99 ms” for Roam. Ookla’s state-level data for the second half of 2025 shows ten states with median multi-server latency under 40 ms; the best sat at 37 ms.

Jitter. This is the number that matters most for voice and the one Starlink does not publish. Geoff Huston’s APNIC analysis measured an average jitter of 6.7 ms between successive round-trip samples on an idle link. The same work saw handover spikes adding roughly 30 to 50 ms, with a worst case jump from 30 to 80 ms. APNIC’s NANOG 93 presentation describes handovers adding “very short-term additional queuing latency of about 80 to 100 ms”.

Packet loss. APNIC’s 1-second paced pings found a long-term average loss a little over 1 percent, in a 1 to 2 percent range unrelated to congestion. The NANOG material puts “micro-loss” at about 1.4 percent, mostly from handover events.

Upload. Ookla’s 2025 Global Satellite Broadband Performance Report put the US median upload at 16.91 Mbps (download 117.74 Mbps); a separate Ookla H2 2025 cut reported 19.3 Mbps up and 133.8 Mbps down.

Latency is a non-issue and average jitter is excellent. What can hurt a call is short loss bursts at handovers and jitter caused by your own traffic filling the uplink.

Why Starlink switches satellites every 15 seconds, and what it does to a call

A Starlink satellite travels at about 27,000 km/h and crosses your dish’s field of view in minutes, so the terminal must keep changing satellites. The scheduler assigns satellites in 15-second slots, globally synchronized, so every terminal reconfigures at the same instants. FCC filings quoted in the 2023 paper “A Multifaceted Look at Starlink Performance” and APNIC’s measurements both show it.

For a voice stream sending a packet every 20 ms, a reconfiguration means a handful of packets arrive late or not at all and round-trip time jumps for a fraction of a second, then settles until the next boundary. Three mechanisms absorb this.

The jitter buffer. Every client holds incoming audio briefly before playing it, so a packet that arrives 40 ms late still plays in order. Every millisecond of buffer is a millisecond of added mouth-to-ear delay. A 60 ms buffer on a 40 ms link gives roughly 100 ms one-way, still under the G.114 guideline. Adaptive buffers grow with variation and shrink when the link is calm; on Starlink they settle slightly larger than on fiber.

Packet loss concealment and forward error correction. Opus, the codec in WebRTC softphones, Teams and most cloud phone systems, scales from 6 to 510 kbps and includes in-band FEC: perceptually important frames are re-encoded at a lower bitrate and tucked into the following packet, so a single lost packet can be rebuilt from the next one. The Multifaceted paper observed Zoom sending more FEC traffic over Starlink than over wired links for this reason. G.711 at 64 kbps and G.729 at 8 kbps, the codecs on most SIP desk phones, have simpler concealment.

Short duration. A sub-second disturbance every 15 seconds means the link is clean more than 95 percent of the time. What is audible is a disturbance landing on a queue that is already full, which brings us to the uplink.

Why upload is the real constraint for Starlink voice calls

A single audio call is tiny: Cisco puts one at 21 to 320 kbps depending on codec; Teams recommends 58 kbps; Zoom lists 60 to 80 kbps. Against 17 to 19 Mbps of median upload you could in theory run a hundred calls. But voice is symmetric and everything else is not: a screen share, a cloud backup, a security camera or a colleague’s 1080p video call (3.8 Mbps up on Zoom) all compete for the same upstream lane, and when it fills, packets queue.

Bufferbloat on the Starlink terminal. Bufferbloat is high latency that appears when other traffic is on the network, caused by excessive buffering in the modem. The Multifaceted paper measured Starlink round-trip times rising two to four times, to roughly 400 to 500 ms, during bulk downloads, and APNIC attributes handover spikes to “deep buffers” in the user terminal.

The fix is Smart Queue Management: a router that keeps its own queue short and feeds the terminal slightly slower than it can forward, so the deep buffer never fills. The bufferbloat.net guidance: measure the real link speed, enable SQM (cake or fq_codel) at those values, then adjust until loaded latency stays within 15 to 25 ms of idle.

Which Starlink plan a business should use for VoIP

What differs between plans is priority during congestion, upload range, IP addressing and mobility. Prices are from Starlink’s published US consumer information sheets and service-plan document. SpaceX raised several consumer prices by $5 to $10 per month in May 2026, so confirm at checkout.

Priority is about congestion, not speed. Starlink’s service-plan document states that Priority and Mobile Priority users take precedence over Standard, and Standard over Mobile, and that higher-priority users “see faster, more consistent speeds during peak use”. If your team calls in the evening, when residential cells are busiest, Priority data is the difference between predictable and unpredictable jitter.

A public IP removes a class of SIP problems. Most Starlink plans sit behind carrier-grade NAT (CGNAT), and the stock router cannot forward ports. Priority plans let you enable a public IPv4 from the business dashboard; it is dynamic but tends to stay stable for long periods.

Priority data runs out. Voice itself uses little (a G.711 call at 64 kbps per direction is about 29 MB per hour each way, before packet overhead), but once the allowance is consumed the plan falls to Standard priority or bills $0.74 per GB. Size the allowance for the site’s total traffic, not for the calls.

Does Starlink Mini work for VoIP?

Yes, with expectations set correctly. The Mini is a 1.1 kg flat panel with Wi-Fi 5 (802.11ac), a built-in router, a 110-degree field of view and a 25 to 40 W draw. Starlink’s Roam sheet lists the Mini kit at $599 and Roam performance of 30 to 100 Mbps down, 10 to 30 Mbps up, latency under 99 ms.

For a field rep making one call at a time, that is plenty. The constraints: Roam is the lowest priority tier, the service may not be used while the dish is in motion, and the wired port needs the Starlink Plug accessory.

How to configure the network for VoIP over Starlink

How to troubleshoot choppy calls on Starlink

Step 1: Read the Starlink app statistics. Open the app, tap the network card, then Statistics (or Advanced). Latency should sit in the 25 to 60 ms band and, more importantly, stay flat. Ping drop rate should be under 1 percent; anything above that is audible. The obstruction map needs 6 to 12 hours of satellite passes to populate, so leave the dish in place for a day before judging it.

Step 2: Fix obstructions before anything else. Under 1 percent obstruction is the target for calls; 0.5 to 2 percent can still interrupt them; above 5 percent, move the dish.

Step 3: Run a continuous ping from a wired machine. Ping a stable target (your provider’s SIP server or a public resolver) once per second for ten minutes with no other traffic. Then repeat while uploading a large file. If latency jumps into the hundreds of milliseconds under load, you have bufferbloat: lower the upload shaping rate until loaded ping stays within about 25 ms of idle.

Step 4: Check the call-quality metrics your phone system already collects. Cloud phone systems, Teams and Zoom report per-call jitter, loss, RTT and a MOS estimate. MOS is the 1 (bad) to 5 (excellent) listening-quality scale; ITU reference values put G.711 at 4.1 and G.729 at 3.92, so a call near 4.0 is as good as the codec allows. MOS that drops only during large uploads is queuing; only at certain hours is congestion (a Priority plan is the lever); only on Wi-Fi softphones is the wireless link.

Step 5: Compare codecs and buffer settings. Switch softphones from G.729 to Opus if you can; its in-band FEC is built for short, bursty loss. If a desk phone exposes its jitter buffer, set it to adaptive with a maximum around 80 to 100 ms, which covers the handover spikes APNIC measured while keeping one-way delay under 150 ms.

Step 6: Rule out double NAT and SIP ALG. One-way audio, calls that drop shortly after answer, or phones that register and then go silent are NAT symptoms, not satellite symptoms. Confirm bypass mode is actually on (your router’s WAN interface should hold the address Starlink assigns, not a private address from the Starlink router) and that SIP ALG is off.

Sources

• Starlink says SpaceX targeting 2026 for launch of Starship-ready terabit satellites (The Register, reporting the Starlink 2025 network update)
• Starlink Residential Consumer Information Sheet (PDF)
• Starlink Priority Consumer Information Sheet (PDF)
• Starlink Roam Consumer Information Sheet (PDF)
• Starlink Service Plans (PDF)
• Starlink median U.S. speeds still short of federal minimums: Ookla report (Telecompetitor)
• Starlink nearing full FCC broadband compliance: Ookla report (Telecompetitor)
• Starlink Hits New Highs in the US (Benton Institute summary of Ookla)
• Starlink broadband speeds increase in 2025, but they’re still not great (Fierce Network)
• Starlink Price Increase: Monthly Plans Rise $5-$10 (5Gstore)
• Here’s How Much Starlink Will Cost You (SlashGear)
• Starlink Bypass Mode Explained (Yahoo Tech / BGR)
• Guide: Starlink Port Forwarding and Workarounds (DishyTech)
• Starlink app guide (SatSpeedCheck)
• A Transport Protocol’s View of Starlink (Geoff Huston, APNIC ISP Column, May 2024)
• Starlink presentation at NANOG 93 (APNIC Labs, February 2025, PDF)
• A Multifaceted Look at Starlink Performance (arXiv 2310.09242)
• ITU-T Recommendation G.114: One-way transmission time (PDF)
• Cisco Press: QoS Requirements of Voice
• Understand Teams network requirements (Microsoft Learn)
• Zoom Video SDK system requirements (bandwidth table)
• RFC 6716: Definition of the Opus Audio Codec
• Voice Codecs: G.711, G.729 and Opus in Practice (SIPPulse)
• What’s toll-quality voice? (Computerworld, MOS reference values)
• What can I do about Bufferbloat? (bufferbloat.net)

All sources accessed October 19, 2026.