📞 VoIP Bandwidth Calculator

How VoIP bandwidth is actually calculated

A codec's advertised bitrate is only the voice payload. Every packet that carries it also drags along an RTP header (12 bytes), a UDP header (8 bytes) and an IP header (20 bytes for IPv4) — 40 bytes of overhead before the packet even reaches the wire, where Ethernet adds ~18 more.

The formula this tool uses is:

packets/sec   = 1000 / ptime(ms)
payload bytes = codec_bitrate × ptime / 8000
total bytes   = payload + IP + UDP + RTP + L2 + tunnel
bandwidth     = total bytes × 8 × packets/sec

This is why G.729 is not eight times cheaper than G.711. At 20 ms packetization, G.711 carries 160 bytes of audio per packet and G.729 carries 20 — but both pay the same 40-byte header tax, 50 times a second. G.711 lands near 80 kbps and G.729 near 24 kbps at the IP layer: a 3.3× saving, not 8×.

Why packetization (ptime) matters so much

Raising ptime from 20 ms to 40 ms halves the packet rate, so you pay the header tax half as often. The trade-off is delay and loss sensitivity: each packet now holds twice as much audio, so losing one punches a bigger hole in the conversation, and you add packetization delay to the end-to-end budget. 20 ms is the near-universal default because it balances those; 30–40 ms is a reasonable choice on constrained links.

Sizing a SIP trunk

Multiply the per-call figure by your peak concurrent calls, not your average, and size the link for the busiest hour. Two practical cautions: bandwidth is full duplex, so check that your uplink — usually the smaller number on an asymmetric connection — can carry the one-direction figure for every simultaneous call. And leave headroom: signalling (SIP messages themselves), call recording, and anything else sharing the link all compete with the audio.

If calls are crossing a VPN, add the tunnel overhead above — IPsec in tunnel mode can add ~56 bytes per packet, which on G.729 at 20 ms is more than double the bandwidth.

Frequently asked questions

How much bandwidth does a G.711 call use?

About 87 kbps in each direction over Ethernet at 20 ms packetization (80 kbps measured at the IP layer), so roughly 174 kbps for the full duplex conversation. Ten concurrent G.711 calls need close to 0.9 Mbps in each direction.

Is G.729 worth the licence and quality loss?

On a constrained WAN link, yes — it cuts per-call bandwidth to roughly a third. On a modern fibre or business broadband connection the saving rarely justifies the narrowband audio quality, and G.711 (or G.722 for HD) is the better default.

What is cRTP and should I enable it?

Compressed RTP squeezes the 40-byte IP/UDP/RTP header down to 2–4 bytes, which is dramatic for low-bitrate codecs. It only works hop by hop on a link where both ends support it — typically a private WAN or leased line. It cannot survive a path across the public internet, so don't plan an internet-facing trunk around it.

Does this include SIP signalling?

No — these figures are the RTP media stream only. SIP signalling is small and bursty (call setup and teardown rather than continuous), so it's normally covered by the headroom you should already be leaving rather than calculated per call.

Related reading: what NAT means for SIP and VoIP calls, VoIP terminology explained, and why VoIP calls have no audio.