Difference between High Density Spectral(HDS) and Spectrogram Analysis

While we disclosed our HDS technology on extracting muscular movement information from human voice, some of the audience thought we are applying multiple filters on spectrogram to extract “muscular movement information”. This is a common misunderstanding. We need to clarify the difference between them.

With classical signal processing, applying filters would remove unwanted signals for inspection, it doesn’t produce more information. And according to the mathematical frequency response, there is limitation of the passing bank width, hence, it would include the influence of neighboring frequency sources. This would make spectogram would only be able to extract the source with relatively highest amplitude. Also, in order to raise bank resolution, a long time duration would be required to complete the extraction. This limits the time and frequency response sensitivity.?

This limitation preventing the conventional technology to resolve the muscular movements during pronunciation. The muscles action moves within 0.1s to perform the action required for each phoneme, and among different muscular movement source groups, they may only have a frequency difference less than 20Hz.?

As a result, the conventional sound analysis only able to resolve the detail of long, strong and stationary phonemes in reality, which are vowels, or prolonged consonants.

With High Density Spectral(HDS) data, the resolution is over 100 times that of those from conventional tools, and NO FILTER is required to separate different signal sources. As in such resolution, each source is clearly isolated from their neighboring signals, and time length of samples required to complete such calculation is less than 0.01s in such detail.?

In the example showing a piece of human speech being analyzed in a conventional spectrogram. Applying filters over different bands range would be like slicing the spectrogram accordingly. No additional information is obtained from filtering.

With HDS, the change is isolated into each source, and being able to capture the change of frequency across a large range, which is never observed with conventional spectrograms. And those changes in frequency existed in reality, such as the characteristic of breaking glass sound or metal collusion sounds . Those information can be the missing rings among all signal analysis.

Back to the case of human speech, the detail level from HDS is far over the requirement of muscular movement during pronunciation. This is the reason for the detailed breakdown of muscular movement in human speech only available from us.?

Filtering is still a key technology in signal processing for selecting a specific range of data and performing regional changes, but no more in analyzing tasks with HDS.