Slashdot reports that Microsoft has received a patent for the Automatic Censorship of Audio Data for Broadcast link here. It sends us to the patent itself which describes "methods for muting offensive words" or making them unintelligible or replacing them with "less offensive words." A word or syllable is rejected when there is a probability above a threshold that it meets of test of "offensiveness" link here.

I ask myself if MS is serious with this patent. The probability of false positives or altering meaning is high so that I can see damage suits arising. I also read the patent as being a hunting license rather than an existing device and ask myself why it was granted. The patent itself notes that a delay in transmission with humans reviewing the speech to determine offense has generally worked. Is this MS in its latest guise as patent troll?

The Patent Office seems to have lost all sense.

Microsoft was not the first. Peter Vogel had published patent applications as far back as 1989 in Australia and Great Britain for automatic censorship programs. Indeed, Microsoft is a Johnny-come-lately to the censorship game, because others have already covered this territory, without the dynamic variation in threshold claimed by Microsoft.

You are correct that such a system will likely give false positives and false negatives. It might be interesting academically and may form the basis for other inventions, but I wonder whether this mechanism is practical?

As for the U.S. patent office losing all sense, they are only the latest in a series of patent offices that "lost sense" with respect to censorship systems, which began in this case with Australia, then Great Britain, then the European Patent Office and finally the United States. Remember that the patent office makes no judgment on the value of an invention; it only determines whether the invention is novel based on published prior art of which it is aware and whether the invention might have utility - even if some people do not agree with the utility.

At least this system should be easy to test for accuracy. Just feed it a) a sermon from a little white nun from Idaho and b) a sample of George Carlin's broadcasts.

The rate of false positives can be estimated from the proportion of beeping in the output from the former.

The rate of false negatives can be estimated from the proportion of non-beeping in the output from the latter.