I manage the speech analytics program and we’ve recently migrated a large group of agents to a new BYOC-Premises trunk. Since the cutover, our ‘Agent Script Compliance’ topic detection accuracy has plummeted.
I dug into the recordings and realized that the audio files for this specific trunk are being processed as a single Mono channel instead of dual-channel Stereo. Because the agent and customer voices are mixed together, the speech engine is frequently attributing the agent’s mandatory compliance phrases to the customer, completely throwing off our scores. Is there a setting on the Edge or the Trunk to force dual-channel recording for analytics purposes?
I’ve seen this happen when the Edge is forced to conserve DSP resources. The recording engine on the Edge captures audio in dual-channel by default because the RTP streams are naturally separate (Tx and Rx).
However, if the Trunk is configured with a ‘Media’ setting that forces ‘Media Bypass’ or if you have a legacy SBC that is proxying the media and mixing it before it reaches the Edge, you’ll lose the stereo separation. Check your SBC configuration first. If the SBC is acting as a B2BUA and bridging the media locally to save WAN bandwidth, it might be mixing the streams.
We run into this in APAC all the time with older carrier links.
Another thing to check is your ‘Recording Policy’ in the Genesys Cloud Admin UI. There is a specific checkbox for ‘Dual Channel’ recording. While it’s usually on by default for Voice, if someone cloned an older policy or set it up via the API without specifying the dual-channel flag, it will default to mixing it down to save storage space. Verify the policy first before you go tearing apart the Edge and SBC network configs!