Topics in DSP (Dog Signal Processing)

Grayson D. Abbott
Center for Advanced Neo-Information and Nonstandard Engineering (CANINE)
Moose Mountain, New Hampshire

The following article summarizes current problems in the field of Dog Signal Processing. The author does not claim a complete understanding of any of these fields, but someone's got to write review articles, right?


Work in Dog Cascade Filters centers around the following problem: Ginger is awakened by the cycling of the refrigerator and begins to bark. Molly, awakened by Ginger's barking, also begins to bark. The neighbors' dogs, depending on certain threshold criteria, may also begin to bark. The neighbors' neighbors' dogs will then bark, etc. In most cases the barking animal density function (BAD) diverges. Adequate filtering to prevent dog cascading is essential.

An even worse problem occurs when Ginger, satisfied that the refrigerator poses no threat, becomes interested in the barking of Molly and the neighbors' dogs - and barks back. In this case we need to apply the recursive forms of the filters.


The need for good bark recognition is not a pressing one; after all, who cares what the dog wants to say ? On the other hand, many researchers prefer this field over human speech recognition because it is easier. First, the vocabulary is highly restricted. The dog only wants to say things like "There's someone at the door" or "I need to go outside" or "I'm hungry"; at least that's all we think the dog should want to say and that's good enough for us. The second advantage of working on bark recognition is that we don't have to "train" the recognition system to the dog's bark - it is easier to train the dog.

The training procedure involves setting up a simple microphone and computer system with a three-light display. The bark recognition system can base its discrimination on any one of several features: formant structure, pitch, or loudness. When the dog barks in a certain way a light will indicate that the dog want to "go outside". Another light will mean that the dog wants to eat. A third light indicates danger at the door. If the appropriate action is taken with each light (i.e. letting the dog out, feeding the dog, yelling "Foo ! Go Away !" at the door) the dog will begin to understand what each bark means and will use them to get what he wants. (This last assertion is controversial. Some psychologists prefer to believe that the association works the other way - the dog begins to want what he gets, and barking in a certain way will cause the dog to suddenly need to go outside.)


Related to the problem of Bark Recognition is that of Barker Recognition. The importance of Barker Recognition can be shown by another example. Ginger, a mature dog, may safely be left inside, alone. Molly, the puppy, exhibits destructive doggy demon processes in the absence of tight Master Control, and, therefore, must be left in the background, or backyard. During long trips by the Master, however, even the mature dog needs a break, and a solution has been proposed, based on Maxwell's Demon, where an automatic door would allow a particular dog to enter and exit freely while excluding all others, bringing a new meaning to "Dog Security Systems".

The key problem is that of recognizing the "safe dog". While some are attempting to solve this problem with methods involving "paw-print" analysis or robot vision systems that count spots, the barker recognition approach seems to hold the greatest promise, due to its simplicity. Just as bark recognition can be based on simple forms of analysis, barker recognition is again a matter of training the barker to produce the right stimulus for the desired response. While in high-security circles it is noted that undesired dog may learn to imitate the welcome dog (referred to as "dog aliasing"), leading researchers feel this is not currently a problem, as the puppy "probably won't get it".


Communicating our wishes to the dog can often be frustrating (see "Programming the User Friendly Dog", Cottrell 1984). While part of the problem lies in the restricted vocabulary (instruction set) of the dog, DSP researchers feel that another part of the problem is phonetic. Dogs simply do not understand English, especially when they don't want to.

Useful devices are now being developed to communicate to the dog in his own language. A hand-held device, including a small speaker and a keyboard marked with dog phonemes, has been developed which can produce sounds that any dog will immediately understand. Useful phrases, such as "I'm happy to see you !", "Let's play !", and "Get away ! This is MY food !", can be produced at the touch of a button.

While this approach proves to be wonderfully useful, the acoustic method is now being augmented with tactile stimulation, as researchers have shown that many dogs experience temporary hearing-loss at certain times of the day.


The application of phased-barker arrays for military purposes is classified at the present time and cannot be discussed here.


This article originally appeared in rec.humor and comp.dsp, on the USENET News. Later, it was printed in Connection Science.

Back to Gray Abbott's Home Page.