Home/Blog/Two-channel recording
Recording · 5 min read

Two-channel recording: the cheapest upgrade for fair QA

Who said it matters as much as what was said. Separate channels settle it.

In short
  • Two-channel (stereo) recording puts the agent and the customer on separate tracks, so nobody has to guess who said what.
  • Scores get fairer, promises are attributed to the right person and talk-time metrics become real.
  • Check your files with Audacity or one ffprobe command, then switch it on in one queue first.
  • Your mono archive is still worth analyzing.

Most phone systems can record a call in one of two ways. In a mono recording, both sides are mixed into a single track. In a two-channel recording, also called stereo, dual-channel or split recording, the agent is on one channel and the customer on the other.

The difference in cost is modest: an uncompressed two-channel file is about twice the size of a mono one, and compressed formats narrow the gap further. The difference for analysis is large, and it shows up exactly where QA is most sensitive.

Why it matters: who said it

Almost every useful finding depends on the speaker. “I'm going to cancel” means one thing from a customer and another from an agent reading back an order. “I'll call you back on Friday” is a promise only if the agent said it. A required disclosure counts only if the agent said it, not if the customer asked about it.

With a mono recording, software has to work out who is speaking from the voices and the words. This is called speaker diarization. It works most of the time and fails in exactly the cases that matter: short replies (“yes”, “okay”), interruptions, people talking over each other, similar voices and hold music.

With two channels, there is nothing to guess. Each sentence comes from a known side of the call.

What gets better

  • Fairer scores. Agents are not marked down for words the customer said, or credited for words they didn't say.
  • Reliable promises. Commitments are attributed to the agent who made them, so follow-up lists are right.
  • Real talk-time metrics. Talk ratio, silence, interruptions and speaking rate can be measured per side.
  • Cleaner transcripts. Overlapping speech is transcribed per speaker instead of being mixed into one garbled line.
  • Stronger evidence. When a finding quotes the customer, you know the quote came from the customer's channel.

An example: the same moment, two ways

Consider a customer interrupting an agent who is reading back cancellation terms:

Mono, after speaker guessingTwo channels
Speaker A: “...so the cancellation fee applies if you, wait, I was told there's no fee, cancel before the 15th.”Agent: “...so the cancellation fee applies if you cancel before the 15th.”
Customer (overlapping): “Wait, I was told there's no fee.”
Mono versus two-channel recording In mono, agent and customer speech is mixed on one track and the speaker has to be guessed. In a two-channel recording, the agent is on one track and the customer on the other, including where they overlap. Mono: one mixed track Two channels: one track per side Who said “wait, I was told there's no fee”? Software has to guess. AGENT CUSTOMER Overlap: still two clean lines Each sentence comes from a known side of the call.
The same moment recorded two ways. Two channels remove the speaker guess, including where people talk over each other.

In the mono version, a scorecard item like “agent stated the cancellation terms clearly” is unclear, and the customer's objection, which might signal a billing dispute, is buried in the agent's sentence. In the two-channel version, both are obvious.

One detail: which channel is the agent?

Two channels do not always put the agent on the same side. Depending on the phone system and the call direction, the agent can be on the left channel in one call and the right in the next. A good system checks this per call instead of assuming. In the VoiceVera case study, the agent's channel was detected for each of the 873 calls, because it changed from call to call.

How to check what you record today

You don't need to ask IT. Take a few recordings from your archive and check the number of channels:

  • In an audio editor such as the free Audacity: open the file. A two-channel recording shows two waveforms, one above the other, labelled left and right.
  • With ffprobe (part of the free FFmpeg tools), run the command below. A value of 2 means two channels.
ffprobe -v error -show_entries stream=channels -of csv=p=0 recording.wav
Two channels is not always two speakersSome systems save a stereo file with the same mixed audio on both channels. If the left and right channels sound identical, the recording is effectively mono. Listen to each side separately, or let the analysis check it for you.

How to switch it on

Most modern phone systems support it; the setting is usually called dual-channel, stereo or split recording.

  • Twilio: request dual-channel recordings when you start recording a call.
  • 3CX, Asterisk, FreePBX: check the recording format setting; stereo output is commonly available, sometimes per queue.
  • Aircall and other hosted platforms: look for “dual-channel” or “stereo” in recording settings, or ask your provider whether it can be enabled.

Make the change in one queue first and check a few new recordings before rolling it out. Recording notices and consent requirements do not change with the format, but it is a good moment to confirm your notice is in place.

What about your existing mono archive?

It is still worth analyzing. Most findings, such as the reason for the call, the outcome or a product defect, do not depend on knowing the exact speaker of every short reply. Speakers are separated in the analysis, and items that depend on who said something are handled more carefully: in VoiceVera, scorecard items that need a known speaker are marked not applicable on mono calls instead of being guessed.

For new calls, though, flipping this one setting is the cheapest quality upgrade available. It makes every downstream number a little more trustworthy, from agent scores to churn signals.

Common questions

Is stereo the same as two-channel?

In call recording, yes: “stereo”, “dual-channel”, “two-channel” and “split” recording usually mean the same thing, one speaker per channel. The exception is the duplicated-mono file described above.

Does it help transcription accuracy?

Usually. Each channel contains one voice, so overlapping speech no longer has to be untangled, and short replies are not lost under the other speaker. The bigger gain, though, is in attribution: knowing who said each sentence.

What about conference calls and transfers?

With three or more people, one channel may hold more than one voice, for example two agents after a warm transfer. Two channels still separate your side from the customer's, which is what most QA items need.

Do we need to re-record or convert old calls?

No. Existing mono recordings can be analyzed as they are. Turn on two-channel recording for new calls, and the share of two-channel calls grows on its own.

VoiceVera works with mono and two-channel recordings. See integrations for supported phone systems.