There are three ways that you can improve restoration results:
Based on your subjective judgment, you can set, modify, or cancel certain detection alarms - see instructions on editing binary files. For mono recordings, detection alarms are displayed in the upper part of the detection window, labeled 1. For stereo soundfiles, two binary signals, labeled 1 and 3, are displayed showing detection alarms for the left and right channels. After introducing changes, choose one of the local ReTouch options from the Restore menu and the signal will be processed following your instructions - the internal outlier detector is switched off when the system is operated in the ReTouch mode.
Remember not to mark blocks wider than 100 samples, because large blocks can be handled differently (see below). Finally, remember that given an unmodified detection file, ReTouch will simply reproduce results yielded earlier by DeClick Plus (unless the smoothing and/or postfiltering factors are altered).
When processing is complete, a Select source dialog appears on the screen, allowing you to compare different files: the source soundfile (original), the destination soundfile prior to retouching (before processing) and the modified destination soundfile (after processing). Use the PLAY
button situated inside the dialog to listen to (locally) the compared soundfiles. Based on listening tests, you can accept the results of retouching (Accept), reject changes introduced by retouching (Reject) or resume editing a detection file (ReEdit). If you choose ReEdit the detection file will be displayed in yellow - this is to remind you that the editing task has not been completed. The white color will be restored once you make a final decision, i.e. accept or reject the changes.
NOTICE
When modifying the upper part of the detection file, you have to be careful. Whenever blocks wider than 20 samples are created there is always a risk that the two-sided reconstruction yielded by the Kalman smoother will suffer from the so-called “covariance defect” - the reconstructed signal will sort of collapse in the middle of the gap. This negative effect may be strongly emphasized for weakly correlated signals.
In order to remove some disturbances, it is necessary to cut and replace large portions of the recording, namely, fragments measured in hundreds of samples rather than in tens. Very large dropouts or scratches fall into this category, as well as some long-lasting chirp and buzz artifacts. When used in the automatic detection (DeClick Plus) mode, the system is not capable of removing such disturbances simply because it is not authorized to question such a large number of samples in a row (otherwise it could potentially start behaving in an unpredictable manner). However, DART XP Pro allows you to replace up to 1500 samples in a row when operating in the ReTouch mode. All you have to do is mark, by editing the lower part of the corresponding detection file (or signals labeled 2 and 4 - for a stereo soundfile), blocks of samples selected for replacement. When ReTouch is used, DART XP Pro will fill each block with the material which resembles either the one immediately preceding the gap (left-hand processing) or the one succeeding the gap (right-hand processing). The system usually does a very good job replacing large blocks - and does it in seconds!
When processing is complete, a Select source dialog appears on the screen, allowing you to compare different files: the source soundfile (original), the destination soundfile prior to retouching (before processing), the destination soundfile modified by left-hand processing (after processing [left]) and the destination soundfile modified by right-hand processing (after processing [right]). Use the PLAY
button situated inside the dialog to listen to (locally) the compared soundfiles. Based on listening tests you can accept the results of left-hand or right-hand retouching (Accept) or reject changes introduced by retouching (Reject). If you are not satisfied with the results, choose the ReEdit button and modify the block, making it larger or smaller, or placing it differently. Chances are the new results will be better. And remember: filling large gaps is a VERY DIFFICULT task.
The reconstruction technique used for processing large blocks is entirely different from the one employed in the auto detection mode. It is free from the covariance defect and you don't have to be very careful when choosing the position and size of the block - if the signal is 'stationary enough' (i.e. its local characteristics do not vary significantly over time), it usually does not matter whether you schedule a block of one hundred, two hundreds or even of one thousand samples for reconstruction.
NOTICE
DART XP Pro will not perform the large block reconstruction if any of the created blocks contains more than 1500 samples. Similarly, it will refuse to do the right-hand reconstruction if the spacing between the adjacent blocks is insufficient. Even though there is no such limitation for left-hand reconstruction (a special 'eating one's tail' technique was applied to relax this restriction) one should not abuse this opportunity (just imagine the snake which eats up it's entire tail).
A number of long-lasting disturbances can be removed by local signal processing using the tools from the DART XP Pro Toolbox menu.
Crackles (series of small clicks), often resulting in buzz-like artifacts, can be eliminated by lowpass filtering (a local use of DeHiss with an appropriately chosen gain yields similar effects) - see Lowpass filter .
Low-frequency thumps, typical of large scratches, can be removed by highpass filtering with the cutoff frequency set in the range <50Hz,200Hz> - see Highpass filter .