Why VoIP calls sound bad — and how to tell which problem you have

“The calls sound bad” is the least useful bug report in telephony, because at least five distinct faults produce complaints that sound identical when relayed second-hand. Learning to tell them apart takes about ten minutes and saves weeks of blaming the wrong supplier.

Choppy or broken audio: packet loss

Words drop out mid-sentence, as if someone is talking through a fan. This is packet loss — audio packets that never arrived, leaving gaps the codec cannot fill.

The cause is almost always the network between the two endpoints, not the carrier. Saturated upstream bandwidth is the usual culprit: someone uploading a large file on the same connection is enough. Failing hardware, overloaded Wi-Fi and cheap switches produce the same effect.

How to confirm: place the same call over a wired connection with nothing else running. If it clears up, the problem is local capacity, not routing.

Robotic or metallic voices: jitter

Speech is intelligible but sounds synthetic and unstable. This is jitter — packets arriving in inconsistent order or with variable delay, forcing the jitter buffer to guess.

Jitter frequently comes from Wi-Fi, VPN tunnels that were sized for file transfer rather than real-time media, or a route with too many hops. It is one of the more common side effects of adding encryption without configuring for voice.

Silence in one direction: one-way audio

You can hear them; they cannot hear you, or vice versa. This is nearly always NAT or firewall configuration — signalling negotiated the call successfully, but the RTP media stream is being blocked in one direction.

It is the classic symptom of a firewall configured by someone treating voice like ordinary web traffic. SIP negotiates on one port and then moves audio to a different range entirely; a generic rule set lets the handshake through and drops the audio.

Talking over each other: latency

Both parties keep interrupting because the gap between speaking and being heard is too long. Under about 150ms one-way, conversation feels natural. Past roughly 300ms, people start colliding.

Latency comes from physical distance and from route quality. This is where cheap wholesale routes reveal themselves: a call routed through three intermediate carriers to shave a fraction of a penny accumulates delay at every hop. It is a symptom you cannot fix locally — only better routing solves it.

Hearing yourself: echo

Usually acoustic rather than network related. A speakerphone picking up its own output, a headset with the microphone too close to the earpiece, or a cheap handset with poor echo cancellation. Latency makes existing echo far more noticeable, which is why it often appears alongside routing problems.

Missing keypresses: DTMF failure

Callers press options in your IVR and nothing happens, or the wrong branch triggers. This is a signalling mismatch — typically inband DTMF being passed where RFC 2833 is expected, or a carrier in the path mangling the tones.

Worth calling out because it is one of the clearest fingerprints of a poor-quality route. Tier-1 termination handles DTMF reliably; heavily resold grey routes frequently do not.

A diagnostic order that works

1. Reproduce the fault and note the exact symptom — not “bad”, but which of the six above.
2. Test the same call on a wired connection with no other traffic. This isolates local network problems immediately.
3. Test a different destination. If only one destination is affected, it is a routing problem, not your network.
4. Test at a different time of day. Load-dependent faults point at capacity; constant faults point at configuration.
5. Only then involve your carrier — and give them the SIP trace, the timestamps and the specific numbers.

Steps one to four cost nothing and resolve or correctly attribute most complaints. Skipping them is why so many businesses change provider and find the problem followed them.