Call Stretching
Call stretching occurs when a carrier pads the billed duration beyond when the conversation actually ended. GSNOC detects it with duration distribution analysis, audio tail analysis, and AI classification.
The Call Stretching Problem
Duration padding adds seconds or minutes to billed call lengths after the conversation has ended. It is hard to detect without audio evidence and hard to dispute without statistical proof of systematic behavior.
- Inflated billing from carriers adding seconds to every call on a high-volume route
- Gradual margin erosion that is difficult to attribute to a single cause without systematic analysis
- Hard-to-dispute charges without audio evidence showing the conversation ended before billing stopped
- Statistical baselines obscured when all calls on a route are padded uniformly
How Call Stretching Detection Works
A combination of statistical analysis and audio examination identifies both systematic and opportunistic call stretching.
Call duration distributions are modeled per carrier and route. Anomalous clustering at suspicious durations (e.g. exactly N seconds longer than expected) flags potential padding.
RTP streams are analyzed at the end of flagged calls to detect silence, comfort noise injection, or audio dropout before the BYE message.
A trained model assigns a stretching confidence score combining duration anomaly, audio tail content, carrier history, and billing pattern signals.
Per-carrier stretching rates are tracked over time to distinguish systematic fraud from isolated incidents and to quantify financial impact.
Why GSNOC Call Stretching Detection Is Different
- AI-classified with confidence scores — not just duration thresholds that generate false positives on legitimate long calls
- Audio evidence of silence or comfort noise injection provides disputable proof of padding
- Statistical baseline modeling identifies carriers that stretch all calls uniformly, which naive thresholds miss entirely
Key Metrics
Estimated seconds added to billed duration beyond actual conversation end
AI confidence (0–100) that the duration anomaly represents deliberate padding
Percentage of calls from a carrier classified as stretched over a rolling window
Estimated billing overcharge from stretching across all flagged calls in the analysis period