Describe the returned signal before naming a cause
Choose one affected call and ask which participant hears their own voice, how long after speaking it returns, whether the return is continuous or intermittent and whether it contains speech, room sound, sidetone, tones or media from another participant. Keep echo, normal near-zero-delay handset sidetone, repeated packets, one-way audio and crosstalk as separate symptoms.
The person hearing echo is not automatically the source. Their speech must leave that side and return from a downstream acoustic, electrical or software path. Preserve call direction, endpoint model and firmware, handset or headset, speakerphone state, volume, network path, PBX leg, trunk, codec and timestamps before changing gain or cancellation.
Localize acoustic coupling with one controlled substitution
At the far end of the returning speech path, change from speakerphone to handset at a moderate level without changing the call route. Then substitute one known-good headset or endpoint if authorized. Check speaker and microphone placement, headset seal, duplicated audio devices, room reflections, operating-system monitoring and accessory damage.
If the symptom follows a device, audio mode or level, keep the investigation at that boundary. A disappearing symptom is evidence for the changed path, not proof of a particular defective component. Do not apply blanket microphone or speaker reductions across the fleet; low gain can hide echo while damaging intelligibility.
Separate analog hybrid return from packet-media behavior
When the affected route crosses an ATA, FXO gateway, analog trunk or other two-wire/four-wire boundary, record the exact port, line profile, country setting, impedance, levels and echo-canceller state. Use only the hardware vendor's profile for that model and line; an impedance or gain value copied from another country or interface can create a new problem.
On an all-IP path, compare RTP sequence numbers, timestamps, SSRC and packet arrival at bounded capture points on both PBX legs. A duplicated packet or duplicated decoded stream is different from acoustic speech that was captured again by a microphone. Encrypted media can still show timing and packet identity at a capture point without exposing or proving decoded audio.
Measure delay without treating it as the echo source
Delay makes a returned signal easier to perceive, but delay and the coupling path are separate owners. Keep network round-trip time, RTP arrival variation, jitter-buffer depth, codec packetization and mouth-to-ear delay as different measurements. RFC 3550 interarrival jitter is a smoothed arrival-variation estimate in the RTP clock domain, not a direct echo-delay or audio-quality measurement.
Compare the affected interval on the exact media direction and PBX leg. Queueing, Wi-Fi, VPN, transcoding and an oversized playout buffer can increase delay, yet reducing a buffer does not remove the acoustic or analog return path and may increase late-packet loss.
Use the Asterisk echo test as a bounded media-loop tool
Asterisk's Echo() application sends media and DTMF frames read from the calling channel back to that channel; it does not automatically answer, and # exits. In an authorized diagnostic extension, that loop can help compare outbound and returned media for one endpoint path. It is not an echo canceller, and hearing the deliberate return does not prove where production echo originates.
Restrict the diagnostic route, avoid exposing it to untrusted callers and remove or disable it after the test. Record the application path, answer behavior, endpoint, codec and result. Do not infer DTMF transport correctness, trunk behavior, analog cancellation or another endpoint's acoustic performance from one looped call.
Change one owner and accept double-talk
Change only the demonstrated owner: endpoint placement or volume, accessory, operating-system monitoring, analog line profile, gain, the supported echo canceller, queueing path or buffer policy. Preserve the previous value and a rollback trigger. Avoid stacking endpoint, gateway and PBX processing changes, because improvement cannot then be attributed and competing cancellers can damage speech.
Repeat the original route, endpoint, audio mode and volume, then test quiet speech, loud speech and double-talk where both participants speak. Confirm that echo is below the agreed threshold without clipping, pumping, blocked syllables or one-way audio, and repeat after restart if the control is persistent. No endpoint, accessory, room, analog circuit, gateway, PBX, RTP stream, delay value, canceller, buffer, call or repair was tested for this guide.
Sources & applicability
Primary references for the technical details above. Operational examples and planning checklists are VoIP.info editorial guidance.
- Asterisk Echo application ↗
- IETF RFC 3550 — RTP and RTCP ↗
- Asterisk DAHDI tools system configuration sample ↗
Examples require adaptation to your topology. No live PBX or hardware testing is claimed.