The Self-Deconvolution Data Restoration Algorithm, or SeDDaRA, was developed to quickly and effectively remove degradations such as focus and motion from digital images. However, this technology can and has been applied to one-dimensional signals as well. In fact, our gallery gallery shows several examples of signal processing.
On this page, we will demonstrate our experiements to apply the technology towards recorded acoustic waves, or sound, to remove some artifacts of sound recording and reproduction. These experiments are first-level attempts.
Note: A research paper has been submitted on the restoration of audio-frequency signals using SeDDaRA. See research for details.
A frequency response is imposed on any recording of acoustic waves by the recording medium. In other words, a microphone will alter the way a sound wave sounds because it is difficult to match the response of a microphone to that of the human ear. Better microphones, of course, produce better reproductions. The same can be said for playback. Speakers do not exactly reproduce the sound wave as we first perceived it. The SeDDaRA process can be used to diminish this behavior by correctly the frequency spectrum imposed on the wave by the reproduction system.
The media players on the side have sound clips from many decades ago. The sound was no doubt recorded as an analogue signal, and someone in cyberspace converted them to digital files. We put the clips through our process and produced more 'lifelike' representations.
The first clip is the famous phrase from John F. Kennedy's famous Inaugural speech on January 20, 1961.
The second clip is much older. This is from a speech by Theodore Roosevelt on August 12, 1912.
In any acoustic reproduction system, noise created from electronic amplification can be introduced. We simulated this problem by adding white noise to sound files, and then attempted to restore the sound files using SeDDaRA. Now, SeDDaRA operates like a selective frequency filter, so noise that has the frequency characteristics similar to those in the signal will not be removed.
The third sound file is a clip from a bassoon quartet. Noise was adding random number to each point in the data file. The noise in the deconvolution is significantly reduced, but not totally eliminated.
The last sound wave is of a synthesized flute playing a scale. As before, noise was added using a random number generator. The sound file was restored and is the last media player on the right.
We are very encouraged by these results! Potentially this technology will provide filtering for such devices as hearing aids, sound systems, and telephones with the ability to not only enhance their reproduction, but also diminish the artifacts produced by the instrument that recorded the sound, and the surrounding environment. Exciting stuff!