How it works
Algorithm Rule
The algorithm at the heart of the Supervizer is based on a comparative verification of audio signals. Each audio pair is compared with a reference. A proximity factor measured with a significative difference generates an alert that is reported in the Graphical User Interface.
There is two levels of comparison:
- First, references are compared to the main reference with a substantial acceptance because serious problems need to be reported regardless of the different on-air processing that significantly modify the signal in terms of mathematics.
- other pairs are then compared to their respective reference with alert thresholds a lot more stricts.
Audio Signals verifications are completed in successive tests.The following problems are tested and detected:
- MUTE
- GAIN
- FILTERING
- ERROR
- FALSE SIGNAL
- DELAY
- PROBLEM (undetermined)
If more that one problem is detected on the pair, on the same channel or on different channels (left and right), without any of them sufficiently significant (only the association of the two channels triggered the alert), nor sufficiently predominant over other detected issues, a generic message “PROBLEM” is then displayed. If several problems can be determined with no doubt (each problem would individually trigger an alert), only the first tested problem is reported.
Alerts are temperate by an hysterisis process. Each reported alert is consolidated and checked. This process is applied to every single alert as well as combination of alerts in order to report an alert when different successive issues occur consecutively.
Alert Description
| Displayed alert | Detection | Problem |
|---|---|---|
| MUTE | silence | The maximum level on a window is inferior to the adjusted threshold. Specific case for the main reference: the hysteresis is processed on a longer duration. (“mute_delai” parameter) |
| GAIN | gain difference | The average level of the channel on the bandwidth [fmin_gain, fmax_gain] is superior (resp inferior) in dB to “ecart_gain_max” (resp “ecart_gain_min”) the average level of the reference on the same bandwidth. |
| FILTERING | excessive filtering | The average level of the channel normalized is superior (resp inferior) in dB to “tolerance_gaba x gabarit_max” (resp “tolerance_gaba x gabarit_min”) the average level of the reference on at least 5 bandwidths of the gabarit. |
| ERROR | additive error | The power of the difference between a normalized realigned and its reference is superior in dB to “bruit_max” (in case of a continuous noise for instance) |
| FALSE SIGNAL | inconsistent signal | The similarity value is inferior to “correl_seuil” (if we compare two different stations for instance) |
| OFFSET | time-lag | The offset compared to initial delay (“input_delays”) is superior to “max_shift”. Note: “input_delays” is in samples as it needs to be accurate; “max_shift” is a perceptive parameter in ms. |
| PROBLEM | undetermined problem | Several problems of different types succeeded during the integration period, with a rate superior to the hysteresis threshold “hysteresis_high”, with none of these problems clearly predominant. |
