Implementing Comfort Noise Generation (CNG) and Voice Activity Detection (VAD) Tuning in Genesys Cloud CX

How-To-Articles
1

Implementing Comfort Noise Generation (CNG) and Voice Activity Detection (VAD) Tuning in Genesys Cloud CX

What This Guide Covers

This guide details the configuration of Silence Suppression mechanisms, specifically Voice Activity Detection (VAD) and Comfort Noise Generation (CNG), within Genesys Cloud CX SIP Trunks and Endpoint negotiations. The end result is a telephony architecture that optimizes bandwidth utilization during silent periods without degrading perceived audio quality or introducing artifacts during packet loss events.

Prerequisites, Roles & Licensing

  • Licensing Tier: Genesys Cloud Telephony (Cloud Connector or Direct IP Connectivity). This configuration requires access to SIP Trunk settings which are available on all supported telephony licenses.
  • Granular Permissions: You require the Telephony > Trunks > Edit permission to modify trunk-level codec and signaling behaviors. Access to Telephony > Quality > Call Recording is necessary for post-call analysis of audio quality.
  • OAuth Scopes: If automating this via API, use genesyscloud:telephony:trunk:read and genesyscloud:telephony:trunk:write.
  • External Dependencies: A network infrastructure capable of prioritizing RTP traffic (QoS) is assumed. VAD/CNG tuning will not resolve underlying jitter or packet loss without proper QoS tagging on the edge devices.

The Implementation Deep-Dive

1. Understanding the Architecture of Silence Suppression

Before modifying configurations, you must understand how Genesys Cloud CX handles silence. Unlike legacy on-premises PBX systems where VAD is a distinct hardware feature, cloud telephony relies heavily on codec negotiation during the SIP INVITE exchange.

Voice Activity Detection (VAD) determines when a user stops speaking. When no voice activity is detected, the endpoint or the platform stops sending RTP packets for that duration. This saves bandwidth but creates “dead air” which can be disconcerting to callers. Comfort Noise Generation (CNG) solves this by injecting low-level background noise during silent periods.

In Genesys Cloud CX, CNG and VAD are intrinsically linked to the selected codec. The G.729 codec inherently supports VAD and CNG. The G.711 codec (ULAW/ALAW) generally transmits silence as empty packets or requires specific header configurations to enable suppression. Therefore, tuning is not a simple toggle; it is a selection of bandwidth optimization policies that dictate how the platform negotiates these capabilities during call setup.

The Trap: Many architects assume they can force VAD on G.711 via a UI switch without understanding the impact on latency. In Genesys Cloud CX, enabling aggressive silence suppression on high-fidelity codecs often introduces perceptible delay or robotic artifacts if the network has any jitter. Do not enable VAD/CNG on G.711 unless your network is strictly managed and you are certain the round-trip time will not exceed 150ms during silent periods.

Architectural Reasoning: We prioritize G.729 for WAN links where bandwidth is constrained, as it reduces packet size by approximately 80% compared to G.711 while enabling VAD/CNG natively. For LAN or high-bandwidth connections, we default to G.711 without VAD to maintain the highest possible fidelity. This decision minimizes the risk of “choppy audio” during network congestion, as G.729 is more sensitive to packet loss than G.711.

2. Configuring SIP Trunk Codec Policies

The primary mechanism for controlling VAD/CNG behavior in Genesys Cloud CX is the Codec Policy associated with your SIP Trunks. You must ensure that the negotiation allows the endpoints and the cloud platform to agree on a codec that supports silence suppression.

Navigate to Telephony > Configuration > Codecs. Create or edit a policy that prioritizes G.729 if bandwidth optimization is a requirement, or G.711 ulaw/alaw for high-fidelity requirements.

Configuration Steps:

  1. Select the Trunk configuration responsible for external connectivity.
  2. Assign the Codec Policy to this Trunk.
  3. Ensure that G.729a is listed before G.711 ulaw/alaw if your network link is a shared WAN connection.

If you are using the Genesys Cloud API to manage these configurations, use the following payload structure for updating a codec policy. Note that specific VAD parameters are not exposed as individual JSON fields in the public API; they are implicit based on the codec priority order.

PUT /api/v2/settings/telephony/codecs/{CODEC_POLICY_ID}
{
    "id": "codec_policy_id_12345",
    "name": "WAN-Optimized-Codecs",
    "codecs": [
        {
            "type": "G729a",
            "priority": 1,
            "payloadTypeNumber": 18
        },
        {
            "type": "PCMU",
            "priority": 2,
            "payloadTypeNumber": 0
        },
        {
            "type": "PCMA",
            "priority": 3,
            "payloadTypeNumber": 8
        }
    ],
    "status": "ACTIVE"
}

The Trap: Reordering codecs without considering endpoint capabilities. If your SIP Trunk connects to a legacy carrier that does not support G.729a, forcing it as the top priority will cause call failures or fallback to G.711 unexpectedly, potentially bypassing your VAD strategy. Always verify carrier compatibility before reordering priorities.

Architectural Reasoning: We place G.729a at priority 1 because it enables the platform to negotiate silence suppression automatically. When the caller speaks, the RTP stream is active. When they pause, the platform injects Comfort Noise packets rather than sending silence (zero packets). This prevents the carrier from terminating the call due to lack of heartbeat traffic and keeps the media path open without consuming full bandwidth.

3. Endpoint Configuration and Softphone Behavior

The behavior of VAD/CNG is not solely determined by the Cloud platform; it is also influenced by the endpoint device. The Genesys Cloud Contact Center Mobile (CCM) application and the Genesys Cloud Desktop Client handle codec negotiation differently than hardware IP phones connected via a third-party provider.

For internal agents using the softphone client, you must verify that the network settings allow for UDP RTP traffic to flow freely. If your organization uses a firewall that blocks unknown UDP ports, VAD packets may be dropped, resulting in silence during pauses.

Configuration Steps:

  1. In the Genesys Cloud Contact Center Desktop Client Settings, ensure Audio Quality is set to “Standard” or “High”.
  2. Avoid setting audio quality to “Low” as this forces aggressive compression that may disable CNG features on certain builds of the softphone client.
  3. For hardware endpoints (Polycom, Cisco), check the vendor-specific provisioning for “Silence Suppression” and ensure it is enabled.

The Trap: Assuming the Cloud platform controls endpoint behavior entirely. The Genesys Cloud Desktop Client allows local audio processing settings that can override cloud-side codec negotiations. If an agent disables “Noise Suppression” or “Echo Cancellation” locally, VAD may fail to detect silence correctly, causing the system to transmit continuous background noise as voice data. This creates a false sense of activity and wastes bandwidth.

Architectural Reasoning: We enforce endpoint consistency through Mobile Device Management (MDM) policies where possible. By standardizing the audio configuration across all agent devices, we ensure that VAD detection thresholds are consistent. This prevents scenarios where one agent causes high bandwidth usage while another on the same network segment consumes minimal resources due to local setting discrepancies.

Validation, Edge Cases & Troubleshooting

Edge Case 1: Silent Periods Triggering Call Drops

The Failure Condition: A call disconnects unexpectedly after 30 seconds of silence from one party.
The Root Cause: Some carriers interpret a lack of RTP packets (due to VAD) as a connection termination signal and drop the session. This is common when VAD is enabled but CNG is not properly negotiated or supported by the carrier side.
The Solution: Verify that the SIP Trunk settings on the Genesys Cloud side allow for “Keep Alive” signaling during silent periods. In Genesys Cloud, this is often handled by the platform ensuring CNG packets are sent. If the issue persists, you must disable VAD for that specific Trunk or Codec Policy to force continuous packet transmission during silence.
Validation Method: Use the Genesys Cloud Call Recording feature to review a silent call segment. Check the waveform; if there is no audio data between 0:30 and 1:00, but the call did not drop, VAD is functioning correctly. If the call drops, CNG packets are likely being blocked by an intermediate firewall or dropped by the carrier.

Edge Case 2: Audio Artifacts During Network Jitter

The Failure Condition: Callers hear robotic noises or “chatter” during pauses in conversation.
The Root Cause: Comfort Noise Generation relies on the continuous transmission of low-level packets to maintain the perceived silence level. If network jitter causes these CNG packets to arrive out of order or with high latency, the receiver cannot reconstruct the noise floor correctly. The result is audible gaps or digital artifacts.
The Solution: Switch from G.729 (which relies heavily on VAD/CNG for bandwidth efficiency) to G.711 ULAW for this specific call path if jitter exceeds 50ms. Alternatively, reduce the aggressiveness of silence suppression by adjusting codec parameters in the Trunk configuration where available via support escalation or advanced settings.
Validation Method: Monitor Network Quality Metrics in the Genesys Cloud Analytics dashboard during peak hours. Look for high values in “Jitter” and “Packet Loss”. If jitter correlates with audio artifacts, VAD/CNG tuning is insufficient to compensate for network instability.

Edge Case 3: NAT Traversal Issues with CNG

The Failure Condition: One-way audio where the agent can hear the caller but the caller cannot hear the agent during pauses.
The Root Cause: Network Address Translation (NAT) firewalls may block incoming RTP packets if they do not recognize them as part of an established session. VAD/CNG generates new packets frequently during silent periods. Some stateful firewalls interpret this frequent generation of low-bandwidth traffic as suspicious or stale and drop the packets.
The Solution: Configure STUN (Session Traversal Utilities for NAT) on the SIP Trunk to ensure the correct IP address is advertised in the SDP body. Ensure that UDP ports 10000-20000 are open bidirectionally on the firewall.
Validation Method: Perform a packet capture using Wireshark or similar tools during a test call. Filter for rtp and voip. Inspect the payload type of silence packets. If you see RTP headers but no payload data, VAD is active. If you see dropped TCP/UDP handshake retransmissions, NAT traversal is interfering with CNG packet delivery.

Official References