Using 'Spectral view' to Spot Disturbances

Many disturbances that locally corrupt audio signals, can be spotted by analyzing condensed signal plots - they come out as irregularities of the signal pattern. The same technique can be applied in the frequency domain. If you have problems localizing a short-lived disturbance try analyzing the signal time-frequency spectrum in the vicinity of the trouble area. If any of the spectral graphs stands out from the neighboring ones the corresponding time frame is likely to contain the source of your problem.

As an example consider the following time-frequency spectral graph depicted in the area inhabited by a short-lived, high-pitched buzz artifact.

A quick visual inspection clearly indicates a potential problem area. A closer inspection of the signal (a real archive recording) in the vicinity of the 'ridge' seen in the central part of the 3-D plot reveals the following ugly little creature:

which is actually the source of an audible buzz. Note that the localized disturbance has a form of a low-intensity, high-frequency dither of a very short duration (less than 20 milliseconds) and as such would be extremely difficult to spot by other means. It can be easily removed by means of local lowpass filtering of the signal.

Some additional guidelines useful when looking for disturbances are summarized below:

Even though you can use the spectral analysis technique to localize impulsive disturbances (noise pulses can significantly deform spectral graphs) the time-domain analysis is usually more efficient and yields better results. For this reason we recommend you use the Spectral view tool to localize only those disturbances that ‘survive' declicking.

When looking for disturbances make sure the full signal coverage indicator is lit, otherwise the disturbance may hide in the areas not sampled. To enforce the 100% coverage, use long analysis frames and/or small zoom coefficients.