Saltar al contenido
Tecnología 7 de noviembre de 2026 10 lectura de minutos

WebRTC Agent Audio Problems: A Fix Guide

Choppy, robotic or delayed audio on browser agents? Measure jitter, loss and round-trip time with WebRTC stats, then fix the network, the headset or the PC.

D
7 de noviembre de 2026

Respuesta rápida

Choppy or robotic WebRTC softphone audio is usually late or lost packets. Measure it with the browser's getStats data: jitter, packetsLost and concealedSamples on the inbound audio, and roundTripTime from the remote report. Then fix in order: wired network instead of Wi-Fi, no VPN in the media path, a relay if the network blocks direct paths, a wired headset, and a lighter PC. Compare the agent leg with the carrier leg before blaming either.

An agent says the customer sounds like a robot, or the customer says the agent keeps cutting out. Supervisors hear it on one desk but not the next. Because a browser softphone runs on whatever network, headset and PC the agent has, audio problems are often not a platform fault. They are a measurement problem first.

This guide gives a symptom map, shows which browser statistics to read and what they mean, and then walks through the network, device and PC causes in the order they are usually found. It is written for contact-centre agents on a browser softphone, but the checks apply to any WebRTC calling. For one-way and no-audio faults, see SIP one-way audio troubleshooting.

Symptom map

What the agent hearsCausa probableFirst check
Choppy, robotic, words missingPacket loss or jitter on the networkWired connection; jitter, packetsLost, concealedSamples
Delay, people talk over each otherHigh round-trip time or a deep jitter bufferroundTripTime, jitterBufferDelay, relay path
Echo of own voiceSpeakers feeding the microphoneUse a headset, not laptop speakers
Audio drops at connectICE cannot find a working pathFirewall, relay availability, VPN
No microphonePermission or wrong device selectedBrowser site permission, device selection

Measure first: what getStats tells you

The browser exposes call statistics through the peer connection's getStats method, which returns a promise that resolves with data about the connection or a specific track (MDN, checked November 7, 2026). The W3C WebRTC statistics specification defines the fields. These are the ones that matter for audio:

  • jitter on the inbound audio: packet jitter in seconds, calculated as defined in RFC 3550 section 6.4.1 (W3C WebRTC Statistics). Rising jitter means packets are arriving at uneven intervals.
  • packetsLost: the total number of RTP packets lost. The specification notes it is an estimate and can be negative if more packets are received than were sent.
  • concealedSamples: samples that were replaced with synthesised audio before playout, for example because of lost packets or packets that arrived too late. Compare it with totalSamplesReceived. A growing share of concealed audio is what the agent hears as robotic or glitchy sound.
  • jitterBufferDelay: the sum of the time each sample spends in the jitter buffer. Dividing it by jitterBufferEmittedCount gives the average delay. A buffer that grows to absorb jitter adds delay.
  • roundTripTime on the remote-inbound report: an estimate from RTCP receiver reports, in seconds. It is absent until a receiver report with a non-zero delay value (DLSR) has been received.

A minimal way to read them during a test call, run from the browser console of the page that owns the peer connection (named pc here):

const report = await pc.getStats();
report.forEach((s) => {
  if (s.type === "inbound-rtp" && s.kind === "audio") {
    console.log("jitter (s)", s.jitter);
    console.log("packets lost", s.packetsLost);
    console.log("concealed samples", s.concealedSamples, "of", s.totalSamplesReceived);
  }
  if (s.type === "remote-inbound-rtp" && s.kind === "audio") {
    console.log("round trip (s)", s.roundTripTime);
  }
});

Do not judge a single reading. Sample every few seconds for the length of a call, note when the numbers jump and match that moment against what the agent heard and what else was happening on the network. The thresholds that matter depend on your codec and tolerance, so build your own baseline from good calls and compare.

Network: Wi-Fi, VPNs, ICE and TURN

Wi-Fi versus wired. Wi-Fi shares the air with other devices, and interference can show up as jitter and loss in the pattern above. The simplest test in this guide is also the most useful: put the agent on a cable for one shift and compare their numbers.

VPNs and security software. A VPN can add a long path, an extra hop of encryption and a different route for media from the one the signalling takes. If only agents on the VPN are affected, test with it off, or ask the network team whether media can be excluded from the tunnel.

Uplink saturation. A large upload, a backup or a video meeting on the same connection can fill the uplink and delay the call's packets. Check what else the connection is doing when the numbers jump.

ICE and TURN relays. A WebRTC call starts by finding a working network path using ICE (RFC 8445). If a host is behind a NAT, direct communication can be impossible in some situations, and it then needs an intermediate node acting as a relay; the protocol for that is TURN (RFC 8656). RFC 8656 also notes that many enterprise networks do not permit direct UDP between internal clients and outside addresses. ICE ranks direct candidates above relayed ones by default (RFC 8445, section 5.1.2.2). A relayed path is often longer than a direct one, so a relay can add delay, but without it some agents would have no audio at all. When audio fails only at connect, and only on some networks, check whether those networks block the traffic a relay needs. If your platform operates the relay infrastructure, that is something to confirm with the platform team, not guess.

Devices: headsets, Bluetooth, permission and selection

  • Use a wired headset. A USB headset avoids laptop speaker echo and the radio problems of wireless ones.
  • Be careful with Bluetooth. A Bluetooth headset adds a wireless link that can drop out, and on some headset and operating-system combinations the call audio sounds worse while its microphone is in use. If only Bluetooth agents complain, swap the headset before touching the network.
  • Check the permission. If the browser blocked the microphone for the site, there is no outbound audio at all. Re-allow it in the site settings.
  • Check the selected devices. Operating systems and browsers keep their own default input and output. A call can go to a monitor's speakers or a webcam's microphone without anyone noticing.
  • Remove echo at the source. Laptop speakers feeding the microphone cause echo that software can only partly hide. Headsets fix it.

Browser and CPU load

Encoding and decoding audio is light work, but a PC under pressure can still miss deadlines. Typical culprits are a video meeting running at the same time, dozens of browser tabs, a screen-share, heavy extensions and a machine that is also running a virus scan or an update. If the stats show rising jitter and loss but the network test is clean, look at the machine. Close what is not needed, restart the browser, and test on a different PC to see whether the fault follows the person or the machine.

Media is always encrypted

Unlike plain SIP and RTP, WebRTC media cannot be captured and played back with an ordinary packet trace. RFC 8827 (WebRTC security architecture) says all media channels must be secured with SRTP, that media traffic must not be sent over plain RTP or RTCP, and that DTLS-SRTP must be offered for every media channel (section 6.5). What this means in practice: a capture shows that packets flow, how many, and how often, but not the audio. The browser's own statistics are therefore the right tool for call quality, and the packet capture is useful for confirming that packets are leaving and arriving at all.

Agent leg or carrier leg?

A contact-centre call has at least two legs: from the agent's browser to the platform, and from the platform to the carrier and the customer. A quality problem on one leg is not fixed by changing the other. A quick way to separate them:

  1. Test the agent leg alone. Call an internal number or a test line from the agent's browser, and read the stats. If these calls are clean, the agent's network and device are fine.
  2. Test other agents on the same carrier route. If several agents on different networks hear the same problem to the same destinations, suspect the carrier leg.
  3. Look for a pattern. One agent, all calls: the agent leg. All agents, one destination or carrier: the carrier leg. One time of day: congestion.
  4. Compare with a different carrier route for the same destination, if you have one.

If carrier-side faults are suspected, a SIP trace and the codes in SIP error codes on outbound calls are the next place to look.

Cómo encaja DialerBee

DialerBee Escritorio del agente includes a WebRTC softphone in the browser tab with dial pad, DTMF, mute and hold, with nothing for the agent to install. A media pre-flight checks the microphone, the speakers and the media path before the agent goes ready, so a broken headset is found before a customer is on the line. On the supervisor side, mean opinion score is captured per call and low-MOS alerts surface on the wallboard while the shift is running (see the operations center). A low score tells you which calls to investigate, and the browser stats above tell you why.

For background, What Is a WebRTC Softphone y What Is Jitter cover the terms, WebRTC vs SIP phones compares the two approaches, and browser dialer vs installed software covers the trade-off for IT.

Agent pre-shift audio checklist

  • Wired network connection, not Wi-Fi, where possible.
  • Wired USB headset plugged in before opening the browser.
  • Microphone permission allowed for the dialer site.
  • Correct input and output device selected in the browser and the operating system.
  • Run the media pre-flight and fix any warning before going ready.
  • VPN off for calling, unless IT has confirmed media is routed correctly.
  • Video meetings, backups and large downloads closed or paused.
  • Unneeded browser tabs closed.
  • Make one test call and listen before the first customer call.
  • Tell the supervisor which desk, time and destination if a problem appears, so it can be traced.

Preguntas frecuentes

Why is my browser softphone audio choppy?

Choppy audio usually means packets are arriving late or not at all. The browser reports this in its stats: a rising jitter value, a growing count of lost packets and an increasing number of concealed samples. Weak Wi-Fi, a busy VPN and a saturated uplink are the usual causes, so test on a wired connection first and watch the numbers.

What do jitter, packet loss and round-trip time mean for a call?

Jitter is the variation in packet arrival time, reported in seconds. Packet loss is the count of RTP packets that never arrived. Round-trip time is how long a message takes to go to the other end and back. Jitter and loss cause choppy or robotic sound, while a high round-trip time causes delay and people talking over each other.

How do I measure WebRTC call quality in the browser?

Call getStats on the peer connection and read the inbound-rtp entry for the audio track: jitter, packetsLost, jitterBufferDelay and concealedSamples. The remote-inbound-rtp entry carries the roundTripTime estimated from RTCP reports. Sample it every few seconds during a call and compare values over time rather than looking at a single reading.

Is WebRTC audio encrypted, and can I capture it like normal SIP?

WebRTC media is always encrypted. RFC 8827 says all media channels must be secured with SRTP, that media must not be sent as plain RTP, and that DTLS-SRTP must be offered for every media channel. A packet capture of a browser call therefore shows encrypted media, which is why the browser's own stats are the right measuring tool.

Do my agents need a TURN server?

Some will. A host behind a NAT or a strict firewall may not be able to reach its peer directly, and TURN lets it use an intermediate relay (RFC 8656). ICE (RFC 8445) ranks direct paths above relayed ones by default and uses a relay when no direct path works. If agents on locked-down networks have no audio, a missing or blocked relay is a likely cause.

Why does a Bluetooth headset make the audio worse?

Bluetooth adds a radio link between the headset and the PC, and a weak or interfered link adds dropouts. How a headset carries its microphone over that link also varies by model and operating system, and on some combinations the call audio sounds worse while the microphone is in use. A wired USB headset removes both variables, which is why it is the first thing to try when only some agents are affected.

Fuentes

  1. W3C, Identifiers for WebRTC's Statistics API (jitter, packetsLost, roundTripTime, jitterBufferDelay, concealedSamples). w3.org/TR/webrtc-stats (consultado el 7 de noviembre de 2026)
  2. MDN, RTCPeerConnection: getStats() method. developer.mozilla.org (consultado el 7 de noviembre de 2026)
  3. RFC 8445, Interactive Connectivity Establishment (ICE), section 5.1.2.2. rfc-editor.org/rfc/rfc8445 (consultado el 7 de noviembre de 2026)
  4. RFC 8656, Traversal Using Relays around NAT (TURN), section 1. rfc-editor.org/rfc/rfc8656 (consultado el 7 de noviembre de 2026)
  5. RFC 8827, WebRTC Security Architecture, section 6.5. rfc-editor.org/rfc/rfc8827 (consultado el 7 de noviembre de 2026)

Lecturas relacionadas

¿Listo para ver DialerBee en acción?

Solicite una demostración en vivo de 15 minutos o comience una prueba gratuita de 14 días: sin presentaciones, sin compromiso. Las llamadas empiezan en cuanto su propia troncal SIP o PBX está conectada y aprobada.

Prueba gratuita de 14 días · crédito de IA gratuito · sin tarjeta · 11 idiomas · BYOC · controles que apoyan el cumplimiento

Ver el sitio completo en inglés →