What is a dynamic time deformation?

Dynamic time deformation (DTW) includes a calculation method called an algorithm for comparing sounds, videos and graphics that can be similar, but samples that may have subtle differences. Calculations usually formulate the linear representation of the sample and measure differences as a time function. Different sample elements can be mapped on a grid to identify similarities, while commands for functions often use symbols to identify each variable. For example, speech recognition sometimes uses dynamic time deformations to match words, even when they speak different speeds or certain parts are expressed differently.

Many speech recognition programs use dynamic time deformations because people often speak at different speeds. Some vowel sounds may be reported differently depending on emotions or other factors. Some programs can recognize the words spoken no matter who speaks. For this reason, it is usually not effective to add distances in internal time to compare sounds.With DTW, different time -specific points for each signal are analyzed; These distances are calculated on the grid, which runs from the bottom left to the upper right right. Letters are used to represent changes between one source and the other. The solution of the algorithm is usually a larger number, the more two samples. This concept is often used for speech recognition and spelling control and genetic material analysis.

In some measurements, frequency changes can compensate for dynamic time deformation. The signals can be calculated so that their form is used regardless of frequency. Modulated signals can also be a problem, but a grid that calculates the distances between the line segments instead of points, can compensate.

Sequential alignment is generally mathematical and some computer programming skills are needed for full understanding. Dynamic time deformation algorithmsIt depends on some basic conditions for realistic calculation of differences between sound or visual samples. Since the sample as a path along the grid, the algorithm often follows the rules such as the path, cannot go back and that one step is measured at a time. In addition to the lower left to the right right format, measurements are limited to places near the diagonal line. Values ​​that are too steep or shallow are often ignored because they can cause mistakes in final measurement.